diff --git a/official/cv/yolov5/README.md b/official/cv/yolov5/README.md
index fb8d8b39ce7d89d153f065e62cfdfa192d0e2814..21e5517e464f1ec0591e2d4fe6af7258195aab60 100644
--- a/official/cv/yolov5/README.md
+++ b/official/cv/yolov5/README.md
@@ -41,7 +41,7 @@ Note that you can run the scripts with **COCO2017 **or any other datasets with t
After installing MindSpore via the official website, you can start training and evaluation as follows:
```bash
-#run training example(1p) by python command
+#run training example(1p) on Ascend by python command
python train.py \
--data_dir=xxx/dataset \
--is_distributed=0 \
@@ -56,17 +56,28 @@ python train.py \
```bash
# For Ascend device, distributed training example(8p) by shell script
bash run_distribute_train.sh xxx/dateset/ xxx/cspdarknet.ckpt rank_table_8pcs.json
+
+# For GPU device, distributed training example(8p) by shell script
+bash run_distribute_train_gpu.sh xxx/dateset [RANK_SIZE]
```
```bash
-# run evaluation by python command
+# run evaluation on Ascend by python command
python eval.py \
聽聽聽 --data_dir=xxx/dataset \
-聽聽聽 --testing_shape=640 > log.txt 2>&1 &
+聽聽聽 --eval_shape=640 > log.txt 2>&1 &
+
+# run evaluation on GPU by python command
+python eval.py \
+ --device_target="GPU" \
+ --data_dir=xxx/dataset \
+ --yolov5_version='yolov5s' \
+ --pretrained="***/*.ckpt" \
+ --eval_shape=640 > log.txt 2>&1 &
```
```bash
-# run evaluation by shell script
+# run evaluation on Ascend by shell script
bash run_eval.sh xxx/dataset xxx/yolov5.ckpt
```
@@ -78,14 +89,20 @@ bash run_eval.sh xxx/dataset xxx/yolov5.ckpt
鈹溾攢鈹€ model_zoo
鈹溾攢鈹€ README.md // descriptions about all the models
鈹溾攢鈹€ yolov5
- 鈹溾攢鈹€ README.md // descriptions about yolov5
+ 鈹溾攢鈹€ README.md // descriptions about yolov5
鈹溾攢鈹€ scripts
鈹� 鈹溾攢鈹€run_distribute_train.sh // launch distributed training(8p) in ascend
- 鈹� 鈹溾攢鈹€run_eval.sh // shell script for evaluation
+ 鈹� 鈹溾攢鈹€run_distribute_train_gpu.sh // launch distributed training(8p) in GPU
+ 鈹� 鈹溾攢鈹€run_standalone_train.sh // launch 1p training in ascend
+ 鈹� 鈹溾攢鈹€run_eval.sh // shell script for evaluation
鈹� 鈹溾攢鈹€rank_table_8pcs.json // the example of rank table settings for 8p training
+ 鈹溾攢鈹€model_utils
+ 鈹� 鈹溾攢鈹€config.py // getting config parameters
+ 鈹� 鈹溾攢鈹€device_adapter.py // getting device info
+ 鈹� 鈹溾攢鈹€local_adapter.py // getting device info
+ 鈹� 鈹溾攢鈹€moxing_adapter.py // Decorator
鈹溾攢鈹€ src
- 鈹� 鈹溾攢鈹€config.py // parameter configuration
- 鈹� 鈹溾攢鈹€backbone.py // backbone of network
+ 鈹� 鈹溾攢鈹€backbone.py // backbone of network
鈹� 鈹溾攢鈹€distributed_sampler.py // iterator of dataset
鈹� 鈹溾攢鈹€initializer.py // initializer of parameters
鈹� 鈹溾攢鈹€logger.py // log function
@@ -95,9 +112,10 @@ bash run_eval.sh xxx/dataset xxx/yolov5.ckpt
鈹� 鈹溾攢鈹€util.py // util function
鈹� 鈹溾攢鈹€yolo.py // yolov5 network
鈹� 鈹溾攢鈹€yolo_dataset.py // create dataset for YOLOV5
- 鈹溾攢鈹€ train.py // training script
- 鈹溾攢鈹€ eval.py // evaluation script
- 鈹溾攢鈹€ export.py // export script
+ 鈹溾攢鈹€ default_config.yaml // parameter configuration
+ 鈹溾攢鈹€ train.py // training script
+ 鈹溾攢鈹€ eval.py // evaluation script
+ 鈹溾攢鈹€ export.py // export script
```
## [Script Parameters](#contents)
@@ -108,7 +126,7 @@ Major parameters in train.py are:
optional arguments:
--device_target device where the code will be implemented: "Ascend", default is "Ascend"
- --data_dir Train dataset directory.
+ --data_dir Train dataset directory.
--per_batch_size Batch size for Training. Default: 8.
--pretrained_backbone The ckpt file of CSPDarknet53. Default: "".
--resume_yolov5 The ckpt file of YOLOv5, which used to fine tune.Default: ""
@@ -157,6 +175,8 @@ python train.py \
--lr_scheduler=cosine_annealing > log.txt 2>&1 &
```
+You should fine tune the params when run training 1p on GPU
+
The python command above will run in the background, you can view the results through the file `log.txt`.
After training, you'll get some checkpoint files under the **outputs** folder by default. The loss value will be achieved as follows:
@@ -224,7 +244,7 @@ Before running the command below, please check the checkpoint path used for eval
python eval.py \
聽聽聽 --data_dir=xxx/dataset \
聽聽聽 --pretrained=xxx/yolov5.ckpt \
-聽聽聽 --testing_shape=640 > log.txt 2>&1 &
+聽聽聽 --eval_shape=640 > log.txt 2>&1 &
OR
# run evaluation by shell script
bash run_eval.sh xxx/dataset xxx/yolov5.ckpt
@@ -258,34 +278,34 @@ Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.677
YOLOv5 on 118K images(The annotation and data format must be the same as coco2017)
-| Parameters | YOLOv5 |
-| -------------------------- | ------------------------------------------------------------ |
-| Resource | Ascend 910 锛汣PU 2.60GHz锛�192cores; Memory, 755G |
-| uploaded Date | 7/12/2021 (month/day/year) |
-| MindSpore Version | 1.2.0 |
-| Dataset | 118K images |
-| Training Parameters | epoch=300, batch_size=8, lr=0.02,momentum=0.9,warmup_epoch=20 |
-| Optimizer | Momentum |
-| Loss Function | Sigmoid Cross Entropy with logits, Giou Loss |
-| outputs | boxes and label |
-| Loss | 111.970097 |
-| Speed | 8p about 450 FPS |
-| Total time | 8p 21h28min |
-| Checkpoint for Fine tuning | 53.62M (.ckpt file) |
-| Scripts | https://gitee.com/mindspore/mindspore/tree/master/model_zoo/ |
+| Parameters | YOLOv5s | YOLOv5s |
+| -------------------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| Resource | Ascend 910 锛汣PU 2.60GHz锛�192cores; Memory, 755G | GPU NV SMX2 V100-32G |
+| uploaded Date | 7/12/2021 (month/day/year) | 9/15/2021 (month/day/year) |
+| MindSpore Version | 1.2.0 | 1.3.0 |
+| Dataset | 118K images | 118K images |
+| Training Parameters | epoch=300, batch_size=8, lr=0.02,momentum=0.9,warmup_epoch=20| epoch=300, batch_size=32, lr=0.025, warmup_epoch=20, 8p |
+| Optimizer | Momentum | Momentum |
+| Loss Function | Sigmoid Cross Entropy with logits, Giou Loss | Sigmoid Cross Entropy with logits, Giou Loss |
+| outputs | boxes and label | boxes and label |
+| Loss | 111.970097 | 85 |
+| Speed | 8p about 450 FPS | 8p about 290 FPS |
+| Total time | 8p 21h28min | 8p 35h |
+| Checkpoint for Fine tuning | 53.62M (.ckpt file) | 58.87M (.ckpt file) |
+| Scripts | https://gitee.com/mindspore/mindspore/tree/master/model_zoo/ | https://gitee.com/mindspore/mindspore/tree/master/model_zoo/ |
### Inference Performance
-| Parameters | YOLOv5 |
-| ------------------- | --------------------------- |
-| Resource | Ascend 910 锛汣PU 2.60GHz锛�192cores; Memory, 755G |
-| Uploaded Date | 7/12/2021 (month/day/year) |
-| MindSpore Version | 1.2.0 |
-| Dataset | 20K images |
-| batch_size | 1 |
-| outputs | box position and sorces, and probability |
-| Accuracy | mAP >= 36.7%(shape=640) |
-| Model for inference | 56.67M (.ckpt file) |
+| Parameters | YOLOv5s | YOLOv5s |
+| ------------------- | -----------------------------------------------| ---------------------------------------------|
+| Resource | Ascend 910 锛汣PU 2.60GHz锛�192cores; Memory, 755G | GPU NV SMX2 V100-32G |
+| Uploaded Date | 7/12/2021 (month/day/year) | 9/15/2021 (month/day/year) |
+| MindSpore Version | 1.2.0 | 1.3.0 |
+| Dataset | 20K images | 20K images |
+| batch_size | 1 | 1 |
+| outputs | box position and sorces, and probability | box position and sorces, and probability |
+| Accuracy | mAP >= 36.7%(shape=640) | mAP >= 36.7%(shape=640) |
+| Model for inference | 56.67M (.ckpt file) | 58.87M (.ckpt file) |
### Transfer Learning
diff --git a/official/cv/yolov5/default_config.yaml b/official/cv/yolov5/default_config.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..abbfc9d218e9b470471f144a9b5fe37838f88561
--- /dev/null
+++ b/official/cv/yolov5/default_config.yaml
@@ -0,0 +1,185 @@
+# Builtin Configurations(DO NOT CHANGE THESE CONFIGURATIONS unless you know exactly what you are doing)
+enable_modelarts: False
+# Url for modelarts
+data_url: ""
+train_url: ""
+checkpoint_url: ""
+# Path for local
+output_dir: "/cache"
+data_path: "/cache/data"
+output_path: "/cache/train"
+load_path: "/cache/checkpoint_path"
+device_target: "Ascend"
+need_modelarts_dataset_unzip: True
+modelarts_dataset_unzip_name: "coco"
+
+# ==============================================================================
+# Train options
+data_dir: "/data/coco"
+per_batch_size: 32
+yolov5_version: "yolov5s"
+pretrained_backbone: ""
+resume_yolov5: ""
+pretrained_checkpoint: ""
+
+lr_scheduler: "cosine_annealing"
+lr: 0.013
+lr_epochs: "220,250"
+lr_gamma: 0.1
+eta_min: 0.0
+T_max: 300
+max_epoch: 320
+warmup_epochs: 20
+weight_decay: 0.0005
+momentum: 0.9
+loss_scale: 1024
+label_smooth: 0
+label_smooth_factor: 0.1
+log_interval: 100
+ckpt_path: "outputs/"
+ckpt_interval: -1
+is_save_on_master: 1
+is_distributed: 0
+rank: 0
+group_size: 1
+need_profiler: 0
+training_shape: ""
+resize_rate: 10
+is_modelArts: 0
+
+# Eval options
+pretrained: ""
+log_path: "outputs/"
+ann_val_file: ""
+eval_nms_thresh: 0.6
+eval_shape: ""
+ignore_threshold: 0.7
+test_ignore_threshold: 0.001
+multi_label: True
+multi_label_thresh: 0.1
+
+# Export options
+device_id: 0
+batch_size: 1
+testing_shape: 640
+ckpt_file: ""
+file_name: "yolov5"
+file_format: "MINDIR"
+dataset_path: ""
+ann_file: ""
+
+
+# Other default config
+hue: 0.015
+saturation: 1.5
+value: 0.4
+jitter: 0.3
+
+multi_scale: [[320, 320],
+ [352, 352],
+ [384, 384],
+ [416, 416],
+ [448, 448],
+ [480, 480],
+ [512, 512],
+ [544, 544],
+ [576, 576],
+ [608, 608],
+ [640, 640],
+ [672, 672],
+ [704, 704],
+ [736, 736],
+ [768, 768]]
+num_classes: 80
+max_box: 150
+
+# h->w
+anchor_scales: [[12, 16],
+ [19, 36],
+ [40, 28],
+ [36, 75],
+ [76, 55],
+ [72, 146],
+ [142, 110],
+ [192, 243],
+ [459, 401]]
+
+out_channel: 255 # 3 * (num_classes + 5)
+
+input_shape: [[3, 32, 64, 128, 256, 512, 1],
+ [3, 48, 96, 192, 384, 768, 2],
+ [3, 64, 128, 256, 512, 1024, 3],
+ [3, 80, 160, 320, 640, 1280, 4]]
+
+# test_param
+test_img_shape: [640, 640]
+
+labels: [ 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat',
+ 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat',
+ 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack',
+ 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
+ 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket',
+ 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
+ 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair',
+ 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
+ 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book',
+ 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush' ]
+
+coco_ids: [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27,
+ 28, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 53,
+ 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80,
+ 81, 82, 84, 85, 86, 87, 88, 89, 90 ]
+
+result_files: './result_Files'
+
+---
+
+# Help description for each configuration
+# Train options
+data_dir: "Train dataset directory."
+per_batch_size: "Batch size for Training."
+pretrained_backbone: "The ckpt file of CspDarkNet53."
+resume_yolov5: "The ckpt file of YOLOv5, which used to fine tune."
+pretrained_checkpoint: "The ckpt file of YOLOv5CspDarkNet53."
+lr_scheduler: "Learning rate scheduler, options: exponential, cosine_annealing."
+lr: "Learning rate."
+lr_epochs: "Epoch of changing of lr changing, split with ','."
+lr_gamma: "Decrease lr by a factor of exponential lr_scheduler."
+eta_min: "Eta_min in cosine_annealing scheduler."
+T_max: "T-max in cosine_annealing scheduler."
+max_epoch: "Max epoch num to train the model."
+warmup_epochs: "Warmup epochs."
+weight_decay: "Weight decay factor."
+momentum: "Momentum."
+loss_scale: "Static loss scale."
+label_smooth: "Whether to use label smooth in CE."
+label_smooth_factor: "Smooth strength of original one-hot."
+log_interval: "Logging interval steps."
+ckpt_path: "Checkpoint save location."
+ckpt_interval: "Save checkpoint interval."
+is_save_on_master: "Save ckpt on master or all rank, 1 for master, 0 for all ranks."
+is_distributed: "Distribute train or not, 1 for yes, 0 for no."
+rank: "Local rank of distributed."
+group_size: "World size of device."
+need_profiler: "Whether use profiler. 0 for no, 1 for yes."
+training_shape: "Fix training shape."
+resize_rate: "Resize rate for multi-scale training."
+ann_file: "path to annotation"
+each_multiscale: "Apply multi-scale for each scale"
+labels: "the label of train data"
+multi_label: "use multi label to nms"
+multi_label_thresh: "multi label thresh"
+
+# Eval options
+pretrained: "model_path, local pretrained model to load"
+log_path: "checkpoint save location"
+ann_val_file: "path to annotation"
+
+# Export options
+device_id: "Device id for export"
+batch_size: "batch size for export"
+testing_shape: "shape for test"
+ckpt_file: "Checkpoint file path for export"
+file_name: "output file name for export"
+file_format: "file format for export"
+result_files: 'path to 310 infer result floder'
diff --git a/official/cv/yolov5/eval.py b/official/cv/yolov5/eval.py
index 8be991e01ac24a2ec9f77ba9a3f4b80c13fb6733..76b1d9dbe0275a2d854cd796cac8e938d3e774d4 100644
--- a/official/cv/yolov5/eval.py
+++ b/official/cv/yolov5/eval.py
@@ -14,11 +14,9 @@
# ============================================================================
"""YoloV5 eval."""
import os
-import argparse
import datetime
import time
import sys
-import ast
from collections import defaultdict
import numpy as np
@@ -34,64 +32,36 @@ import mindspore as ms
from src.yolo import YOLOV5
from src.logger import get_logger
from src.yolo_dataset import create_yolo_dataset
-from src.config import ConfigYOLOV5
-
-parser = argparse.ArgumentParser('mindspore coco testing')
-
-# device related
-parser.add_argument('--device_target', type=str, default='Ascend',
- help='device where the code will be implemented. (Default: Ascend)')
-
-# dataset related
-parser.add_argument('--data_dir', type=str, default='/data/coco', help='train data dir')
-parser.add_argument('--per_batch_size', default=1, type=int, help='batch size for per gpu')
-
-# network related
-parser.add_argument('--pretrained', default='', type=str, help='model_path, local pretrained model to load')
-parser.add_argument('--yolov5_version', default='yolov5s', type=str,
- help='The version of YOLOv5, options: yolov5s, yolov5m, yolov5l, yolov5x')
-
-# logging related
-parser.add_argument('--log_path', type=str, default='outputs/', help='checkpoint save location')
-
-# detect_related
-parser.add_argument('--nms_thresh', type=float, default=0.6, help='threshold for NMS')
-parser.add_argument('--ann_file', type=str, default='', help='path to annotation')
-parser.add_argument('--testing_shape', type=str, default='', help='shape for test ')
-parser.add_argument('--ignore_threshold', type=float, default=0.001, help='threshold to throw low quality boxes')
-parser.add_argument('--multi_label', type=ast.literal_eval, default=True, help='whether to use multi label')
-parser.add_argument('--multi_label_thresh', type=float, default=0.1, help='threshhold to throw low quality boxes')
-parser.add_argument('--is_modelArts', type=int, default=0,
- help='Trainning in modelArts or not, 1 for yes, 0 for no. Default: 0')
-
-args, _ = parser.parse_known_args()
-args.rank = 0
-
-if args.is_modelArts:
- args.data_root = os.path.join(args.data_dir, 'val2017')
- args.ann_file = os.path.join(args.data_dir, 'annotations')
+
+from model_utils.config import config
+from model_utils.moxing_adapter import moxing_wrapper
+from model_utils.device_adapter import get_device_id, get_device_num
+
+config.rank = 0
+
+if config.is_modelArts:
+ config.data_root = os.path.join(config.data_dir, 'val2017')
+ config.ann_file = os.path.join(config.data_dir, 'annotations')
import moxing as mox
- local_data_url = os.path.join('/cache/data', str(args.rank))
- local_annFile = os.path.join('/cache/data', str(args.rank))
- local_pretrained = os.path.join('/cache/data', str(args.rank))
+ local_data_url = os.path.join(config.data_path, str(config.rank))
+ local_annFile = os.path.join(config.data_path, str(config.rank))
+ local_pretrained = os.path.join(config.data_path, str(config.rank))
- temp_str = args.pretrained.split('/')[-1]
- args.pretrained = args.pretrained[0:args.pretrained.rfind('/')]
+ temp_str = config.pretrained.split('/')[-1]
+ config.pretrained = config.pretrained[0:config.pretrained.rfind('/')]
- mox.file.copy_parallel(args.data_root, local_data_url)
- args.data_root = local_data_url
+ mox.file.copy_parallel(config.data_root, local_data_url)
+ config.data_root = local_data_url
- mox.file.copy_parallel(args.ann_file, local_annFile)
- args.ann_file = os.path.join(local_data_url, 'instances_val2017.json')
+ mox.file.copy_parallel(config.ann_file, local_annFile)
+ config.ann_file = os.path.join(local_data_url, 'instances_val2017.json')
- mox.file.copy_parallel(args.pretrained, local_pretrained)
- args.pretrained = os.path.join(local_data_url, temp_str)
+ mox.file.copy_parallel(config.pretrained, local_pretrained)
+ config.pretrained = os.path.join(local_data_url, temp_str)
else:
- args.data_root = os.path.join(args.data_dir, 'val2017')
- args.ann_file = os.path.join(
- args.data_dir,
- 'annotations/instances_val2017.json')
+ config.data_root = os.path.join(config.data_dir, 'val2017')
+ config.ann_file = os.path.join(config.data_dir, 'annotations/instances_val2017.json')
class Redirct:
@@ -109,17 +79,8 @@ class DetectionEngine:
"""Detection engine."""
def __init__(self, args_detection):
- self.ignore_threshold = args_detection.ignore_threshold
- self.labels = ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat',
- 'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat',
- 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack',
- 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
- 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket',
- 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
- 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair',
- 'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
- 'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book',
- 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush']
+ self.ignore_threshold = args_detection.test_ignore_threshold
+ self.labels = args_detection.labels
self.num_classes = len(self.labels)
self.results = {}
self.file_path = ''
@@ -128,14 +89,11 @@ class DetectionEngine:
self._coco = COCO(self.ann_file)
self._img_ids = list(sorted(self._coco.imgs.keys()))
self.det_boxes = []
- self.nms_thresh = args_detection.nms_thresh
+ self.nms_thresh = args_detection.eval_nms_thresh
self.multi_label = args_detection.multi_label
self.multi_label_thresh = args_detection.multi_label_thresh
self.coco_catids = self._coco.getCatIds()
- self.coco_catIds = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27,
- 28, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 53,
- 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80,
- 81, 82, 84, 85, 86, 87, 88, 89, 90]
+ self.coco_catIds = args_detection.coco_ids
def do_nms_for_results(self):
"""Get result boxes."""
@@ -335,29 +293,82 @@ def convert_testing_shape(args_testing_shape):
testing_shape = [int(args_testing_shape), int(args_testing_shape)]
return testing_shape
+def modelarts_pre_process():
+ '''modelarts pre process function.'''
+ def unzip(zip_file, save_dir):
+ import zipfile
+ s_time = time.time()
+ if not os.path.exists(os.path.join(save_dir, config.modelarts_dataset_unzip_name)):
+ zip_isexist = zipfile.is_zipfile(zip_file)
+ if zip_isexist:
+ fz = zipfile.ZipFile(zip_file, 'r')
+ data_num = len(fz.namelist())
+ print("Extract Start...")
+ print("unzip file num: {}".format(data_num))
+ data_print = int(data_num / 100) if data_num > 100 else 1
+ i = 0
+ for file in fz.namelist():
+ if i % data_print == 0:
+ print("unzip percent: {}%".format(int(i * 100 / data_num)), flush=True)
+ i += 1
+ fz.extract(file, save_dir)
+ print("cost time: {}min:{}s.".format(int((time.time() - s_time) / 60),
+ int(int(time.time() - s_time) % 60)))
+ print("Extract Done.")
+ else:
+ print("This is not zip.")
+ else:
+ print("Zip has been extracted.")
-if __name__ == "__main__":
+ if config.need_modelarts_dataset_unzip:
+ zip_file_1 = os.path.join(config.data_path, config.modelarts_dataset_unzip_name + ".zip")
+ save_dir_1 = os.path.join(config.data_path)
+
+ sync_lock = "/tmp/unzip_sync.lock"
+
+ # Each server contains 8 devices as most.
+ if get_device_id() % min(get_device_num(), 8) == 0 and not os.path.exists(sync_lock):
+ print("Zip file path: ", zip_file_1)
+ print("Unzip file save dir: ", save_dir_1)
+ unzip(zip_file_1, save_dir_1)
+ print("===Finish extract data synchronization===")
+ try:
+ os.mknod(sync_lock)
+ except IOError:
+ pass
+
+ while True:
+ if os.path.exists(sync_lock):
+ break
+ time.sleep(1)
+
+ print("Device: {}, Finish sync unzip data from {} to {}.".format(get_device_id(), zip_file_1, save_dir_1))
+
+ config.log_path = os.path.join(config.output_path, config.log_path)
+
+@moxing_wrapper(pre_process=modelarts_pre_process)
+def run_eval():
start_time = time.time()
device_id = int(os.getenv('DEVICE_ID')) if os.getenv('DEVICE_ID') else 0
# device_id = 1
- context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target, device_id=device_id)
+ context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target, device_id=device_id)
# logger
- args.outputs_dir = os.path.join(args.log_path, datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
+ config.outputs_dir = os.path.join(config.log_path, datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
rank_id = int(os.getenv('DEVICE_ID', '0'))
- args.logger = get_logger(args.outputs_dir, rank_id)
+ config.logger = get_logger(config.outputs_dir, rank_id)
context.reset_auto_parallel_context()
parallel_mode = ParallelMode.STAND_ALONE
context.set_auto_parallel_context(parallel_mode=parallel_mode, gradients_mean=True, device_num=1)
- args.logger.info('Creating Network....')
+ config.logger.info('Creating Network....')
dict_version = {'yolov5s': 0, 'yolov5m': 1, 'yolov5l': 2, 'yolov5x': 3}
- network = YOLOV5(is_training=False, version=dict_version[args.yolov5_version])
+ network = YOLOV5(is_training=False, version=dict_version[config.yolov5_version])
- args.logger.info(args.pretrained)
- if os.path.isfile(args.pretrained):
- param_dict = load_checkpoint(args.pretrained)
+ config.logger.info(config.pretrained)
+ if os.path.isfile(config.pretrained):
+ param_dict = load_checkpoint(config.pretrained)
param_dict_new = {}
for key, values in param_dict.items():
if key.startswith('moments.'):
@@ -367,32 +378,31 @@ if __name__ == "__main__":
else:
param_dict_new[key] = values
load_param_into_net(network, param_dict_new)
- args.logger.info('load_model {} success'.format(args.pretrained))
+ config.logger.info('load_model %s success', config.pretrained)
else:
- args.logger.info('{} not exists or not a pre-trained file'.format(args.pretrained))
- assert FileNotFoundError('{} not exists or not a pre-trained file'.format(args.pretrained))
+ config.logger.info('%s not exists or not a pre-trained file', config.pretrained)
+ assert FileNotFoundError('{} not exists or not a pre-trained file'.format(config.pretrained))
exit(1)
- data_root = args.data_root
- ann_file = args.ann_file
+ data_root = config.data_root
+ ann_file = config.ann_file
- config = ConfigYOLOV5()
- if args.testing_shape:
- config.test_img_shape = convert_testing_shape(args.testing_shape)
+ if config.testing_shape:
+ config.test_img_shape = convert_testing_shape(config.eval_shape)
- ds, data_size = create_yolo_dataset(data_root, ann_file, is_training=False, batch_size=args.per_batch_size,
+ ds, data_size = create_yolo_dataset(data_root, ann_file, is_training=False, batch_size=config.per_batch_size,
max_epoch=1, device_num=1, rank=rank_id, shuffle=False, config=config)
- args.logger.info('testing shape : {}'.format(config.test_img_shape))
- args.logger.info('total {} images to eval'.format(data_size))
+ config.logger.info('testing shape : %d', config.test_img_shape)
+ config.logger.info('total %d images to eval', data_size)
network.set_train(False)
# init detection engine
- detection = DetectionEngine(args)
+ detection = DetectionEngine(config)
input_shape = Tensor(tuple(config.test_img_shape), ms.float32)
- args.logger.info('Start inference....')
+ config.logger.info('Start inference....')
for image_index, data in enumerate(ds.create_dict_iterator(num_epochs=1)):
image = data["image"].asnumpy()
image = np.concatenate((image[..., ::2, ::2], image[..., 1::2, ::2],
@@ -407,16 +417,20 @@ if __name__ == "__main__":
output_small = output_small.asnumpy()
image_id_ = image_id_.asnumpy()
image_shape_ = image_shape_.asnumpy()
- detection.detect([output_small, output_me, output_big], args.per_batch_size, image_shape_, image_id_)
+ detection.detect([output_small, output_me, output_big], config.per_batch_size, image_shape_, image_id_)
if image_index % 1000 == 0:
- args.logger.info('Processing... {:.2f}% '.format(image_index * args.per_batch_size / data_size * 100))
+ config.logger.info('Processing... {:.2f}% '.format(image_index * config.per_batch_size / data_size * 100))
- args.logger.info('Calculating mAP...')
+ config.logger.info('Calculating mAP...')
detection.do_nms_for_results()
result_file_path = detection.write_result()
- args.logger.info('result file path: {}'.format(result_file_path))
+ config.logger.info('result file path: %s', result_file_path)
eval_result = detection.get_eval_result()
cost_time = time.time() - start_time
- args.logger.info('\n=============coco eval reulst=========\n' + eval_result)
- args.logger.info('testing cost time {:.2f}h'.format(cost_time / 3600.))
+ eval_log_string = '\n=============coco eval result=========\n' + eval_result
+ config.logger.info(eval_log_string)
+ config.logger.info('testing cost time %.2f h', cost_time / 3600.)
+
+if __name__ == "__main__":
+ run_eval()
diff --git a/official/cv/yolov5/export.py b/official/cv/yolov5/export.py
index a228cef97e5d6301a20840dd245006652ba6932d..de04ff4ce3b7aa0a78ec66227f4c17d9617c0836 100644
--- a/official/cv/yolov5/export.py
+++ b/official/cv/yolov5/export.py
@@ -12,7 +12,6 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
-import argparse
import numpy as np
import mindspore
@@ -21,37 +20,34 @@ from mindspore.train.serialization import export, load_checkpoint, load_param_in
from src.yolo import YOLOV5s_Infer
-parser = argparse.ArgumentParser(description='yolov5 export')
-parser.add_argument("--device_id", type=int, default=0, help="Device id")
-parser.add_argument("--batch_size", type=int, default=1, help="batch size")
-parser.add_argument('--yolov5_version', default='yolov5s', type=str,
- help='The version of YOLOv5, options: yolov5s, yolov5m, yolov5l, yolov5x')
+from model_utils.config import config
+from model_utils.moxing_adapter import moxing_wrapper
-parser.add_argument("--testing_shape", type=int, default=640, help="test shape")
-parser.add_argument("--ckpt_file", type=str, required=True, help="Checkpoint file path.")
-parser.add_argument("--file_name", type=str, default="yolov5", help="output file name.")
-parser.add_argument('--file_format', type=str, choices=["AIR", "ONNX", "MINDIR"], default='MINDIR', help='file format')
-parser.add_argument("--device_target", type=str, choices=["Ascend", "GPU", "CPU"],
- default="Ascend", help="device target")
-args = parser.parse_args()
+def modelarts_pre_process():
+ '''modelarts pre process function.'''
+ config.file_name = os.path.join(config.output_path, config.file_name)
-context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)
-if args.device_target == "Ascend":
- context.set_context(device_id=args.device_id)
-if __name__ == "__main__":
- ts_shape = args.testing_shape // 2
+@moxing_wrapper(pre_process=modelarts_pre_process)
+def run_export():
+ context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target)
+ if config.device_target == "Ascend":
+ context.set_context(device_id=config.device_id)
+ ts_shape = config.testing_shape // 2
dict_version = {'yolov5s': 0, 'yolov5m': 1, 'yolov5l': 2, 'yolov5x': 3}
- args.file_name = args.file_name + '_' + args.yolov5_version
+ config.file_name = config.file_name + '_' + config.yolov5_version
- network = YOLOV5s_Infer(args.testing_shape, version=dict_version[args.yolov5_version])
+ network = YOLOV5s_Infer(config.testing_shape, version=dict_version[config.yolov5_version])
network.set_train(False)
- param_dict = load_checkpoint(args.ckpt_file)
+ param_dict = load_checkpoint(config.ckpt_file)
load_param_into_net(network, param_dict)
- input_data = Tensor(np.zeros([args.batch_size, 12, ts_shape, ts_shape]), mindspore.float32)
+ input_data = Tensor(np.zeros([config.batch_size, 12, ts_shape, ts_shape]), mindspore.float32)
- export(network, input_data, file_name=args.file_name, file_format=args.file_format)
+ export(network, input_data, file_name=config.file_name, file_format=config.file_format)
print('==========success export===============')
+
+if __name__ == "__main__":
+ run_export()
diff --git a/official/cv/yolov5/model_utils/__init__.py b/official/cv/yolov5/model_utils/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/official/cv/yolov5/model_utils/config.py b/official/cv/yolov5/model_utils/config.py
new file mode 100644
index 0000000000000000000000000000000000000000..ad0d7497a8e5996a42c54a0cda599f84576fdc8e
--- /dev/null
+++ b/official/cv/yolov5/model_utils/config.py
@@ -0,0 +1,126 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+"""Parse arguments"""
+
+import os
+import ast
+import argparse
+from pprint import pformat
+import yaml
+
+class Config:
+ """
+ Configuration namespace. Convert dictionary to members.
+ """
+ def __init__(self, cfg_dict):
+ for k, v in cfg_dict.items():
+ if isinstance(v, (list, tuple)):
+ setattr(self, k, [Config(x) if isinstance(x, dict) else x for x in v])
+ else:
+ setattr(self, k, Config(v) if isinstance(v, dict) else v)
+
+ def __str__(self):
+ return pformat(self.__dict__)
+
+ def __repr__(self):
+ return self.__str__()
+
+
+def parse_cli_to_yaml(parser, cfg, helper=None, choices=None, cfg_path="default_config.yaml"):
+ """
+ Parse command line arguments to the configuration according to the default yaml.
+
+ Args:
+ parser: Parent parser.
+ cfg: Base configuration.
+ helper: Helper description.
+ cfg_path: Path to the default yaml config.
+ """
+ parser = argparse.ArgumentParser(description="[REPLACE THIS at config.py]",
+ parents=[parser])
+ helper = {} if helper is None else helper
+ choices = {} if choices is None else choices
+ for item in cfg:
+ if not isinstance(cfg[item], list) and not isinstance(cfg[item], dict):
+ help_description = helper[item] if item in helper else "Please reference to {}".format(cfg_path)
+ choice = choices[item] if item in choices else None
+ if isinstance(cfg[item], bool):
+ parser.add_argument("--" + item, type=ast.literal_eval, default=cfg[item], choices=choice,
+ help=help_description)
+ else:
+ parser.add_argument("--" + item, type=type(cfg[item]), default=cfg[item], choices=choice,
+ help=help_description)
+ args = parser.parse_args()
+ return args
+
+
+def parse_yaml(yaml_path):
+ """
+ Parse the yaml config file.
+
+ Args:
+ yaml_path: Path to the yaml config.
+ """
+ with open(yaml_path, 'r') as fin:
+ try:
+ cfgs = yaml.load_all(fin.read(), Loader=yaml.FullLoader)
+ cfgs = [x for x in cfgs]
+ if len(cfgs) == 1:
+ cfg_helper = {}
+ cfg = cfgs[0]
+ cfg_choices = {}
+ elif len(cfgs) == 2:
+ cfg, cfg_helper = cfgs
+ cfg_choices = {}
+ elif len(cfgs) == 3:
+ cfg, cfg_helper, cfg_choices = cfgs
+ else:
+ raise ValueError("At most 3 docs (config, description for help, choices) are supported in config yaml")
+ print(cfg_helper)
+ except:
+ raise ValueError("Failed to parse yaml")
+ return cfg, cfg_helper, cfg_choices
+
+
+def merge(args, cfg):
+ """
+ Merge the base config from yaml file and command line arguments.
+
+ Args:
+ args: Command line arguments.
+ cfg: Base configuration.
+ """
+ args_var = vars(args)
+ for item in args_var:
+ cfg[item] = args_var[item]
+ return cfg
+
+
+def get_config():
+ """
+ Get Config according to the yaml file and cli arguments.
+ """
+ parser = argparse.ArgumentParser(description="default name", add_help=False)
+ current_dir = os.path.dirname(os.path.abspath(__file__))
+ parser.add_argument("--config_path", type=str, default=os.path.join(current_dir, "../default_config.yaml"),
+ help="Config file path")
+ path_args, _ = parser.parse_known_args()
+ default, helper, choices = parse_yaml(path_args.config_path)
+ args = parse_cli_to_yaml(parser=parser, cfg=default, helper=helper, choices=choices, cfg_path=path_args.config_path)
+ final_config = merge(args, default)
+ return Config(final_config)
+
+config = get_config()
diff --git a/official/cv/yolov5/model_utils/device_adapter.py b/official/cv/yolov5/model_utils/device_adapter.py
new file mode 100644
index 0000000000000000000000000000000000000000..7c5d7f837ddaa8f53cf8dc5573cac0e36881e7b1
--- /dev/null
+++ b/official/cv/yolov5/model_utils/device_adapter.py
@@ -0,0 +1,27 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+"""Device adapter for ModelArts"""
+
+from .config import config
+
+if config.enable_modelarts:
+ from .moxing_adapter import get_device_id, get_device_num, get_rank_id, get_job_id
+else:
+ from .local_adapter import get_device_id, get_device_num, get_rank_id, get_job_id
+
+__all__ = [
+ "get_device_id", "get_device_num", "get_rank_id", "get_job_id"
+]
diff --git a/official/cv/yolov5/model_utils/local_adapter.py b/official/cv/yolov5/model_utils/local_adapter.py
new file mode 100644
index 0000000000000000000000000000000000000000..769fa6dc78e59eb66dbc8e6773accdc1d08b649e
--- /dev/null
+++ b/official/cv/yolov5/model_utils/local_adapter.py
@@ -0,0 +1,36 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+"""Local adapter"""
+
+import os
+
+def get_device_id():
+ device_id = os.getenv('DEVICE_ID', '0')
+ return int(device_id)
+
+
+def get_device_num():
+ device_num = os.getenv('RANK_SIZE', '1')
+ return int(device_num)
+
+
+def get_rank_id():
+ global_rank_id = os.getenv('RANK_ID', '0')
+ return int(global_rank_id)
+
+
+def get_job_id():
+ return "Local Job"
diff --git a/official/cv/yolov5/model_utils/moxing_adapter.py b/official/cv/yolov5/model_utils/moxing_adapter.py
new file mode 100644
index 0000000000000000000000000000000000000000..25838a7da99a27a1bb744684c1f75f80f5704688
--- /dev/null
+++ b/official/cv/yolov5/model_utils/moxing_adapter.py
@@ -0,0 +1,116 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+"""Moxing adapter for ModelArts"""
+
+import os
+import functools
+from mindspore import context
+from .config import config
+
+_global_sync_count = 0
+
+def get_device_id():
+ device_id = os.getenv('DEVICE_ID', '0')
+ return int(device_id)
+
+
+def get_device_num():
+ device_num = os.getenv('RANK_SIZE', '1')
+ return int(device_num)
+
+
+def get_rank_id():
+ global_rank_id = os.getenv('RANK_ID', '0')
+ return int(global_rank_id)
+
+
+def get_job_id():
+ job_id = os.getenv('JOB_ID')
+ job_id = job_id if job_id != "" else "default"
+ return job_id
+
+def sync_data(from_path, to_path):
+ """
+ Download data from remote obs to local directory if the first url is remote url and the second one is local path
+ Upload data from local directory to remote obs in contrast.
+ """
+ import moxing as mox
+ import time
+ global _global_sync_count
+ sync_lock = "/tmp/copy_sync.lock" + str(_global_sync_count)
+ _global_sync_count += 1
+
+ # Each server contains 8 devices as most.
+ if get_device_id() % min(get_device_num(), 8) == 0 and not os.path.exists(sync_lock):
+ print("from path: ", from_path)
+ print("to path: ", to_path)
+ mox.file.copy_parallel(from_path, to_path)
+ print("===finish data synchronization===")
+ try:
+ os.mknod(sync_lock)
+ except IOError:
+ pass
+ print("===save flag===")
+
+ while True:
+ if os.path.exists(sync_lock):
+ break
+ time.sleep(1)
+
+ print("Finish sync data from {} to {}.".format(from_path, to_path))
+
+
+def moxing_wrapper(pre_process=None, post_process=None):
+ """
+ Moxing wrapper to download dataset and upload outputs.
+ """
+ def wrapper(run_func):
+ @functools.wraps(run_func)
+ def wrapped_func(*args, **kwargs):
+ # Download data from data_url
+ if config.enable_modelarts:
+ if config.data_url:
+ sync_data(config.data_url, config.data_path)
+ print("Dataset downloaded: ", os.listdir(config.data_path))
+ if config.checkpoint_url:
+ sync_data(config.checkpoint_url, config.load_path)
+ print("Preload downloaded: ", os.listdir(config.load_path))
+ if config.train_url:
+ sync_data(config.train_url, config.output_path)
+ print("Workspace downloaded: ", os.listdir(config.output_path))
+
+ context.set_context(save_graphs_path=os.path.join(config.output_path, str(get_rank_id())))
+ config.device_num = get_device_num()
+ config.device_id = get_device_id()
+ if not os.path.exists(config.output_path):
+ os.makedirs(config.output_path)
+
+ if pre_process:
+ pre_process()
+
+ # Run the main function
+ run_func(*args, **kwargs)
+
+ # Upload data to train_url
+ if config.enable_modelarts:
+ if post_process:
+ post_process()
+
+ if config.train_url:
+ print("Start to copy output directory")
+ sync_data(config.output_path, config.train_url)
+ return wrapped_func
+ return wrapper
diff --git a/official/cv/yolov5/scripts/run_distribute_train.sh b/official/cv/yolov5/scripts/run_distribute_train.sh
index 053c8b73b258c4ca3d90b3c8e643399d04807210..e89e4a427008bc3e88d4d84b1b3b4e31284c13cc 100644
--- a/official/cv/yolov5/scripts/run_distribute_train.sh
+++ b/official/cv/yolov5/scripts/run_distribute_train.sh
@@ -56,7 +56,9 @@ do
rm -rf ./train_parallel$i
mkdir ./train_parallel$i
cp ../*.py ./train_parallel$i
+ cp ../*.yaml ./train_parallel$i
cp -r ../src ./train_parallel$i
+ cp -r ../model_utils ./train_parallel$i
cd ./train_parallel$i || exit
echo "start training for rank $RANK_ID, device $DEVICE_ID"
env > env.log
diff --git a/official/cv/yolov5/scripts/run_distribute_train_gpu.sh b/official/cv/yolov5/scripts/run_distribute_train_gpu.sh
new file mode 100644
index 0000000000000000000000000000000000000000..d16a10af01142bca91984fb4d8811fa60c0a7cda
--- /dev/null
+++ b/official/cv/yolov5/scripts/run_distribute_train_gpu.sh
@@ -0,0 +1,66 @@
+#!/bin/bash
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+if [ $# != 2 ]
+then
+ echo "Usage: sh run_distribute_train_gpu.sh [DATASET_PATH] [RANK_SIZE]"
+exit 1
+fi
+
+get_real_path(){
+ if [ "${1:0:1}" == "/" ]; then
+ echo "$1"
+ else
+ echo "$(realpath -m $PWD/$1)"
+ fi
+}
+
+DATASET_PATH=$(get_real_path $1)
+echo $DATASET_PATH
+
+if [ ! -d $DATASET_PATH ]
+then
+ echo "error: DATASET_PATH=$DATASET_PATH is not a directory"
+exit 1
+fi
+
+export RANK_SIZE=$2
+
+if [ -d "distribute_train" ]; then
+ rm -rf ./distribute_train
+fi
+
+mkdir ./distribute_train
+cp ../*.py ./distribute_train
+cp ../*.yaml ./distribute_train
+cp -r ../src ./distribute_train
+cp -r ../model_utils ./distribute_train
+cd ./distribute_train || exit
+
+mpirun --allow-run-as-root -n $RANK_SIZE --output-filename log_output --merge-stderr-to-stdout \
+nohup python train.py \
+ --device_target=GPU \
+ --per_batch_size=32 \
+ --data_dir=$DATASET_PATH \
+ --is_distributed=1 \
+ --yolov5_version='yolov5s' \
+ --lr=0.025 \
+ --T_max=300 \
+ --max_epoch=300 \
+ --warmup_epochs=20 \
+ --training_shape=640 \
+ --lr_scheduler=cosine_annealing > log.txt 2>&1 &
+cd ..
diff --git a/official/cv/yolov5/scripts/run_eval.sh b/official/cv/yolov5/scripts/run_eval.sh
index 600690de2bb2109a4d6290818fadb74148d05b97..c88f7a2138e4fb02f9c739e855b405b94e18b37c 100644
--- a/official/cv/yolov5/scripts/run_eval.sh
+++ b/official/cv/yolov5/scripts/run_eval.sh
@@ -55,7 +55,9 @@ then
fi
mkdir ./eval
cp ../*.py ./eval
+cp ../*.yaml ./eval
cp -r ../src ./eval
+cp -r ../model_utils ./eval
cd ./eval || exit
env > env.log
echo "start inferring for device $DEVICE_ID"
@@ -63,5 +65,5 @@ python eval.py \
--data_dir=$DATASET_PATH \
--yolov5_version='yolov5s' \
--pretrained=$CHECKPOINT_PATH \
- --testing_shape=640 > log.txt 2>&1 &
+ --eval_shape=640 > log.txt 2>&1 &
cd ..
diff --git a/official/cv/yolov5/scripts/run_standalone_train.sh b/official/cv/yolov5/scripts/run_standalone_train.sh
index 0d7d8bf7c76b8f52b8381e37e16c774c6c3b410e..39a7dc0521e603e0c15057d4beb9b14871f6106b 100644
--- a/official/cv/yolov5/scripts/run_standalone_train.sh
+++ b/official/cv/yolov5/scripts/run_standalone_train.sh
@@ -50,7 +50,9 @@ then
fi
mkdir ./train
cp ../*.py ./train
+cp ../*.yaml ./train
cp -r ../src ./train
+cp -r ../model_utils ./train
cd ./train || exit
echo "start training for device $DEVICE_ID"
env > env.log
diff --git a/official/cv/yolov5/src/config.py b/official/cv/yolov5/src/config.py
deleted file mode 100644
index 3414afa90cd84fc808341065b0e680ac896983cd..0000000000000000000000000000000000000000
--- a/official/cv/yolov5/src/config.py
+++ /dev/null
@@ -1,73 +0,0 @@
-# Copyright 2021 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ============================================================================
-"""Config parameters for yolov5 models."""
-
-
-class ConfigYOLOV5:
- """
- Config parameters for the yolov5.
-
- Examples:
- ConfigYOLOV5()
- """
- # train_param
- # data augmentation related
- hue = 0.015
- saturation = 1.5
- value = 0.4
- jitter = 0.3
-
- resize_rate = 10
- multi_scale = [
- [320, 320],
- [352, 352],
- [384, 384],
- [416, 416],
- [448, 448],
- [480, 480],
- [512, 512],
- [544, 544],
- [576, 576],
- [608, 608],
- [640, 640],
- [672, 672],
- [704, 704],
- [736, 736],
- [768, 768]]
- num_classes = 80
- max_box = 150
-
- # confidence under ignore_threshold means no object when training
- ignore_threshold = 0.7
-
- # h->w
- anchor_scales = [(12, 16),
- (19, 36),
- (40, 28),
- (36, 75),
- (76, 55),
- (72, 146),
- (142, 110),
- (192, 243),
- (459, 401)]
- out_channel = 3 * (num_classes + 5)
-
- input_shape = [[3, 32, 64, 128, 256, 512, 1],
- [3, 48, 96, 192, 384, 768, 2],
- [3, 64, 128, 256, 512, 1024, 3],
- [3, 80, 160, 320, 640, 1280, 4]]
-
- # test_param
- test_img_shape = [640, 640]
diff --git a/official/cv/yolov5/src/yolo.py b/official/cv/yolov5/src/yolo.py
index 245566fe5f98253b372650b9c7e517e1345cf7d5..4071e16464118cbbadc26f304d5b786e449f24d6 100644
--- a/official/cv/yolov5/src/yolo.py
+++ b/official/cv/yolov5/src/yolo.py
@@ -25,15 +25,15 @@ from mindspore.ops import functional as F
from mindspore.ops import composite as C
from src.backbone import YOLOv5Backbone, Conv, BottleneckCSP
-from src.config import ConfigYOLOV5
from src.loss import ConfidenceLoss, ClassLoss
+from model_utils.config import config as default_config
class YOLO(nn.Cell):
def __init__(self, backbone, shape):
super(YOLO, self).__init__()
self.backbone = backbone
- self.config = ConfigYOLOV5()
+ self.config = default_config
self.conv1 = Conv(shape[5], shape[4], k=1, s=1)
self.CSP5 = BottleneckCSP(shape[5], shape[4], n=1*shape[6], shortcut=False)
@@ -117,7 +117,7 @@ class DetectionBlock(nn.Cell):
Args:
scale: Character.
- config: ConfigYOLOV5, Configuration instance.
+ config: config, Configuration instance.
is_training: Bool, Whether train or not, default True.
Returns:
@@ -127,7 +127,7 @@ class DetectionBlock(nn.Cell):
DetectionBlock(scale='l',stride=32)
"""
- def __init__(self, scale, config=ConfigYOLOV5(), is_training=True):
+ def __init__(self, scale, config=default_config, is_training=True):
super(DetectionBlock, self).__init__()
self.config = config
if scale == 's':
@@ -240,7 +240,7 @@ class YoloLossBlock(nn.Cell):
"""
Loss block cell of YOLOV5 network.
"""
- def __init__(self, scale, config=ConfigYOLOV5()):
+ def __init__(self, scale, config=default_config):
super(YoloLossBlock, self).__init__()
self.config = config
if scale == 's':
@@ -339,7 +339,7 @@ class YOLOV5(nn.Cell):
def __init__(self, is_training, version=0):
super(YOLOV5, self).__init__()
- self.config = ConfigYOLOV5()
+ self.config = default_config
# YOLOv5 network
self.shape = self.config.input_shape[version]
@@ -378,7 +378,7 @@ class YoloWithLossCell(nn.Cell):
def __init__(self, network):
super(YoloWithLossCell, self).__init__()
self.yolo_network = network
- self.config = ConfigYOLOV5()
+ self.config = default_config
self.loss_big = YoloLossBlock('l', self.config)
self.loss_me = YoloLossBlock('m', self.config)
self.loss_small = YoloLossBlock('s', self.config)
diff --git a/official/cv/yolov5/train.py b/official/cv/yolov5/train.py
index 532c56cd40bace811d78c73d65f45c145670820b..e3ef1b8a4b982a33710ce3742b3cc45ca5a4785f 100644
--- a/official/cv/yolov5/train.py
+++ b/official/cv/yolov5/train.py
@@ -15,7 +15,6 @@
"""YoloV5 train."""
import os
import time
-import argparse
import datetime
import mindspore as ms
from mindspore.context import ParallelMode
@@ -32,230 +31,222 @@ from src.util import AverageMeter, get_param_groups
from src.lr_scheduler import get_lr
from src.yolo_dataset import create_yolo_dataset
from src.initializer import default_recurisive_init, load_yolov5_params
-from src.config import ConfigYOLOV5
-ms.set_seed(1)
+from model_utils.config import config
+from model_utils.moxing_adapter import moxing_wrapper
+from model_utils.device_adapter import get_device_id, get_device_num
-parser = argparse.ArgumentParser('mindspore coco training')
-
-# device related
-parser.add_argument('--device_target', type=str, default='Ascend', help='device where the code will be implemented.')
-
-# dataset related
-parser.add_argument('--data_dir', default='/data/coco', type=str, help='Train dataset directory.')
-parser.add_argument('--per_batch_size', default=32, type=int, help='Batch size for Training. Default: 8')
-
-# network related
-parser.add_argument('--yolov5_version', default='yolov5s', type=str,
- help='The version of YOLOv5, options: yolov5s, yolov5m, yolov5l, yolov5x')
-parser.add_argument('--pretrained_backbone', default='', type=str, help='The pretrained file of yolov5. Default: "".')
-parser.add_argument('--resume_yolov5', default='', type=str,
- help='The ckpt file of YOLOv5, which used to fine tune. Default: ""')
-
-# optimizer and lr related
-parser.add_argument('--lr_scheduler', default='cosine_annealing', type=str,
- help='Learning rate scheduler, options: exponential, cosine_annealing. Default: exponential')
-parser.add_argument('--lr', default=0.013, type=float, help='Learning rate. Default: 0.01')
-parser.add_argument('--lr_epochs', type=str, default='220,250',
- help='Epoch of changing of lr changing, split with ",". Default: 220,250')
-parser.add_argument('--lr_gamma', type=float, default=0.1,
- help='Decrease lr by a factor of exponential lr_scheduler. Default: 0.1')
-parser.add_argument('--eta_min', type=float, default=0., help='Eta_min in cosine_annealing scheduler. Default: 0')
-parser.add_argument('--T_max', type=int, default=300, help='T-max in cosine_annealing scheduler. Default: 320')
-parser.add_argument('--max_epoch', type=int, default=300, help='Max epoch num to train the model. Default: 320')
-parser.add_argument('--warmup_epochs', default=20, type=float, help='Warmup epochs. Default: 0')
-parser.add_argument('--weight_decay', type=float, default=0.0005, help='Weight decay factor. Default: 0.0005')
-parser.add_argument('--momentum', type=float, default=0.9, help='Momentum. Default: 0.9')
-
-# loss related
-parser.add_argument('--loss_scale', type=int, default=1024, help='Static loss scale. Default: 1024')
-parser.add_argument('--label_smooth', type=int, default=0, help='Whether to use label smooth in CE. Default:0')
-parser.add_argument('--label_smooth_factor', type=float, default=0.1,
- help='Smooth strength of original one-hot. Default: 0.1')
-
-# logging related
-parser.add_argument('--log_interval', type=int, default=100, help='Logging interval steps. Default: 100')
-parser.add_argument('--ckpt_path', type=str, default='outputs/', help='Checkpoint save location. Default: outputs/')
-parser.add_argument('--ckpt_interval', type=int, default=None, help='Save checkpoint interval. Default: None')
-
-parser.add_argument('--is_save_on_master', type=int, default=1,
- help='Save ckpt on master or all rank, 1 for master, 0 for all ranks. Default: 1')
-
-# distributed related
-parser.add_argument('--is_distributed', type=int, default=0,
- help='Distribute train or not, 1 for yes, 0 for no. Default: 1')
-parser.add_argument('--rank', type=int, default=0, help='Local rank of distributed. Default: 0')
-parser.add_argument('--group_size', type=int, default=1, help='World size of device. Default: 1')
-
-# roma obs
-parser.add_argument('--train_url', type=str, default="", help='train url')
-# profiler init
-parser.add_argument('--need_profiler', type=int, default=0,
- help='Whether use profiler. 0 for no, 1 for yes. Default: 0')
-
-# reset default config
-parser.add_argument('--training_shape', type=str, default="", help='Fix training shape. Default: ""')
-parser.add_argument('--resize_rate', type=int, default=10, help='Resize rate for multi-scale training. Default: None')
-parser.add_argument('--is_modelArts', type=int, default=0,
- help='Trainning in modelArts or not, 1 for yes, 0 for no. Default: 0')
-
-args, _ = parser.parse_known_args()
-
-if args.lr_scheduler == 'cosine_annealing' and args.max_epoch > args.T_max:
- args.T_max = args.max_epoch
-
-args.lr_epochs = list(map(int, args.lr_epochs.split(',')))
-
-if args.is_modelArts:
- args.data_root = os.path.join(args.data_dir, 'train2017')
-
- args.annFile = os.path.join(args.data_dir, 'annotations')
- outputs_dir = os.path.join('/cache', args.ckpt_path)
-else:
- args.data_root = os.path.join(args.data_dir, 'train2017')
- args.annFile = os.path.join(
- args.data_dir, 'annotations/instances_train2017.json')
- outputs_dir = args.ckpt_path
-
-deviced = int(os.getenv('DEVICE_ID', '0'))
-context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target,
- save_graphs=False, device_id=deviced)
-# init distributed
-if args.is_distributed:
- if args.device_target == "Ascend":
- init()
+ms.set_seed(1)
- else:
- init("nccl")
- args.rank = get_rank()
- args.group_size = get_group_size()
+def set_default():
+ if config.lr_scheduler == 'cosine_annealing' and config.max_epoch > config.T_max:
+ config.T_max = config.max_epoch
-args.rank_save_ckpt_flag = 0
-if args.is_save_on_master:
- if args.rank == 0:
- args.rank_save_ckpt_flag = 1
-else:
- args.rank_save_ckpt_flag = 1
+ config.lr_epochs = list(map(int, config.lr_epochs.split(',')))
-# logger
-args.outputs_dir = os.path.join(outputs_dir, datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
-args.logger = get_logger(args.outputs_dir, args.rank)
-args.logger.save_args(args)
+ if config.is_modelArts:
+ config.data_root = os.path.join(config.data_dir, 'train2017')
+ config.annFile = os.path.join(config.data_dir, 'annotations')
+ outputs_dir = os.path.join(config.outputs_dir, config.ckpt_path)
+ else:
+ config.data_root = os.path.join(config.data_dir, 'train2017')
+ config.annFile = os.path.join(config.data_dir, 'annotations/instances_train2017.json')
+ outputs_dir = config.ckpt_path
+
+ device_id = get_device_id()
+ context.set_context(mode=context.GRAPH_MODE, enable_auto_mixed_precision=True, device_target=config.device_target,
+ save_graphs=False, device_id=device_id)
+ # init distributed
+ if config.is_distributed:
+ if config.device_target == "Ascend":
+ init()
+ else:
+ init("nccl")
+ config.rank = get_rank()
+ config.group_size = get_group_size()
+
+ config.rank_save_ckpt_flag = 0
+ if config.is_save_on_master:
+ if config.rank == 0:
+ config.rank_save_ckpt_flag = 1
+ else:
+ config.rank_save_ckpt_flag = 1
+ # logger
+ config.outputs_dir = os.path.join(outputs_dir, datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
+ config.logger = get_logger(config.outputs_dir, config.rank)
+ config.logger.save_args(config)
def convert_training_shape(args_training_shape):
training_shape = [int(args_training_shape), int(args_training_shape)]
return training_shape
-
-loss_meter = AverageMeter('loss')
-
-if args.is_modelArts:
- import moxing as mox
- local_data_url = os.path.join('/cache/data', str(args.rank))
- local_annFile = os.path.join('/cache/data', str(args.rank))
- mox.file.copy_parallel(args.data_root, local_data_url)
- args.data_root = local_data_url
-
- mox.file.copy_parallel(args.annFile, local_annFile)
- args.annFile = os.path.join(local_data_url, 'instances_train2017.json')
-
-context.reset_auto_parallel_context()
-parallel_mode = ParallelMode.STAND_ALONE
-degree = 1
-if args.is_distributed:
- parallel_mode = ParallelMode.DATA_PARALLEL
- degree = get_group_size()
-context.set_auto_parallel_context(parallel_mode=parallel_mode, gradients_mean=True, device_num=degree)
-
-dict_version = {'yolov5s': 0, 'yolov5m': 1, 'yolov5l': 2, 'yolov5x': 3}
-network = YOLOV5(is_training=True, version=dict_version[args.yolov5_version])
-# default is kaiming-normal
-default_recurisive_init(network)
-load_yolov5_params(args, network)
-
-network = YoloWithLossCell(network)
-config = ConfigYOLOV5()
-
-config.label_smooth = args.label_smooth
-config.label_smooth_factor = args.label_smooth_factor
-
-if args.training_shape:
- config.multi_scale = [convert_training_shape(args.training_shape)]
-if args.resize_rate:
- config.resize_rate = args.resize_rate
-
-ds, data_size = create_yolo_dataset(image_dir=args.data_root, anno_path=args.annFile, is_training=True,
- batch_size=args.per_batch_size, max_epoch=args.max_epoch,
- device_num=args.group_size, rank=args.rank, config=config)
-
-args.logger.info('Finish loading dataset')
-
-args.steps_per_epoch = int(data_size / args.per_batch_size / args.group_size)
-
-if not args.ckpt_interval:
- args.ckpt_interval = args.steps_per_epoch
-
-lr = get_lr(args)
-
-opt = Momentum(params=get_param_groups(network), momentum=args.momentum, learning_rate=Tensor(lr),
- weight_decay=args.weight_decay, loss_scale=args.loss_scale)
-
-network = TrainingWrapper(network, opt, args.loss_scale // 2)
-network.set_train()
-
-if args.rank_save_ckpt_flag:
- # checkpoint save
- ckpt_max_num = args.max_epoch * args.steps_per_epoch // args.ckpt_interval
- ckpt_config = CheckpointConfig(save_checkpoint_steps=args.ckpt_interval, keep_checkpoint_max=1)
- save_ckpt_path = os.path.join(args.outputs_dir, 'ckpt_' + str(args.rank) + '/')
- ckpt_cb = ModelCheckpoint(config=ckpt_config, directory=save_ckpt_path, prefix='{}'.format(args.rank))
- cb_params = _InternalCallbackParam()
- cb_params.train_network = network
- cb_params.epoch_num = ckpt_max_num
- cb_params.cur_epoch_num = 1
- run_context = RunContext(cb_params)
- ckpt_cb.begin(run_context)
-
-old_progress = -1
-t_end = time.time()
-data_loader = ds.create_dict_iterator(output_numpy=True, num_epochs=1)
-
-for i, data in enumerate(data_loader):
- images = data["image"]
- input_shape = images.shape[2:4]
- images = Tensor.from_numpy(images)
- batch_y_true_0 = Tensor.from_numpy(data['bbox1'])
- batch_y_true_1 = Tensor.from_numpy(data['bbox2'])
- batch_y_true_2 = Tensor.from_numpy(data['bbox3'])
- batch_gt_box0 = Tensor.from_numpy(data['gt_box1'])
- batch_gt_box1 = Tensor.from_numpy(data['gt_box2'])
- batch_gt_box2 = Tensor.from_numpy(data['gt_box3'])
- input_shape = Tensor(tuple(input_shape[::-1]), ms.float32)
- loss = network(images, batch_y_true_0, batch_y_true_1, batch_y_true_2, batch_gt_box0, batch_gt_box1,
- batch_gt_box2, input_shape)
- loss_meter.update(loss.asnumpy())
-
- if args.rank_save_ckpt_flag:
- # ckpt progress
- cb_params.cur_step_num = i + 1 # current step number
- cb_params.batch_num = i + 2
- ckpt_cb.step_end(run_context)
-
- if i % args.log_interval == 0:
- time_used = time.time() - t_end
- epoch = int(i / args.steps_per_epoch)
- fps = args.per_batch_size * (i - old_progress) * args.group_size / time_used
- if args.rank == 0:
- args.logger.info('epoch[{}], iter[{}], {}, fps:{:.2f} imgs/sec, '
- 'lr:{}'.format(epoch, i, loss_meter, fps, lr[i]))
- t_end = time.time()
- loss_meter.reset()
- old_progress = i
-
- if (i + 1) % args.steps_per_epoch == 0 and args.rank_save_ckpt_flag:
- cb_params.cur_epoch_num += 1
-
-if args.is_modelArts:
- mox.file.copy_parallel(src_url='/cache/outputs/', dst_url='obs://hit-cyf/yolov5_npu/outputs/')
-args.logger.info('==========end training===============')
+def modelarts_pre_process():
+ '''modelarts pre process function.'''
+ def unzip(zip_file, save_dir):
+ import zipfile
+ s_time = time.time()
+ if not os.path.exists(os.path.join(save_dir, config.modelarts_dataset_unzip_name)):
+ zip_isexist = zipfile.is_zipfile(zip_file)
+ if zip_isexist:
+ fz = zipfile.ZipFile(zip_file, 'r')
+ data_num = len(fz.namelist())
+ print("Extract Start...")
+ print("unzip file num: {}".format(data_num))
+ data_print = int(data_num / 100) if data_num > 100 else 1
+ i = 0
+ for file in fz.namelist():
+ if i % data_print == 0:
+ print("unzip percent: {}%".format(int(i * 100 / data_num)), flush=True)
+ i += 1
+ fz.extract(file, save_dir)
+ print("cost time: {}min:{}s.".format(int((time.time() - s_time) / 60),
+ int(int(time.time() - s_time) % 60)))
+ print("Extract Done.")
+ else:
+ print("This is not zip.")
+ else:
+ print("Zip has been extracted.")
+
+ if config.need_modelarts_dataset_unzip:
+ zip_file_1 = os.path.join(config.data_path, config.modelarts_dataset_unzip_name + ".zip")
+ save_dir_1 = os.path.join(config.data_path)
+
+ sync_lock = "/tmp/unzip_sync.lock"
+
+ # Each server contains 8 devices as most.
+ if get_device_id() % min(get_device_num(), 8) == 0 and not os.path.exists(sync_lock):
+ print("Zip file path: ", zip_file_1)
+ print("Unzip file save dir: ", save_dir_1)
+ unzip(zip_file_1, save_dir_1)
+ print("===Finish extract data synchronization===")
+ try:
+ os.mknod(sync_lock)
+ except IOError:
+ pass
+
+ while True:
+ if os.path.exists(sync_lock):
+ break
+ time.sleep(1)
+
+ print("Device: {}, Finish sync unzip data from {} to {}.".format(get_device_id(), zip_file_1, save_dir_1))
+
+ config.ckpt_path = os.path.join(config.output_path, config.ckpt_path)
+
+@moxing_wrapper(pre_process=modelarts_pre_process)
+def run_train():
+ set_default()
+ loss_meter = AverageMeter('loss')
+
+ if config.is_modelArts:
+ import moxing as mox
+ local_data_url = os.path.join(config.data_path, str(config.rank))
+ local_annFile = os.path.join(config.data_path, str(config.rank))
+ mox.file.copy_parallel(config.data_root, local_data_url)
+ config.data_root = local_data_url
+
+ mox.file.copy_parallel(config.annFile, local_annFile)
+ config.annFile = os.path.join(local_data_url, 'instances_train2017.json')
+
+ context.reset_auto_parallel_context()
+ parallel_mode = ParallelMode.STAND_ALONE
+ degree = 1
+ if config.is_distributed:
+ parallel_mode = ParallelMode.DATA_PARALLEL
+ degree = get_group_size()
+ context.set_auto_parallel_context(parallel_mode=parallel_mode, gradients_mean=True, device_num=degree)
+
+ dict_version = {'yolov5s': 0, 'yolov5m': 1, 'yolov5l': 2, 'yolov5x': 3}
+ network = YOLOV5(is_training=True, version=dict_version[config.yolov5_version])
+ # default is kaiming-normal
+ default_recurisive_init(network)
+ load_yolov5_params(config, network)
+
+ network = YoloWithLossCell(network)
+
+ config.label_smooth = config.label_smooth
+ config.label_smooth_factor = config.label_smooth_factor
+
+ if config.training_shape:
+ config.multi_scale = [convert_training_shape(config.training_shape)]
+ if config.resize_rate:
+ config.resize_rate = config.resize_rate
+
+ ds, data_size = create_yolo_dataset(image_dir=config.data_root, anno_path=config.annFile, is_training=True,
+ batch_size=config.per_batch_size, max_epoch=config.max_epoch,
+ device_num=config.group_size, rank=config.rank, config=config)
+
+ config.logger.info('Finish loading dataset')
+
+ config.steps_per_epoch = int(data_size / config.per_batch_size / config.group_size)
+
+ if config.ckpt_interval <= 0:
+ config.ckpt_interval = config.steps_per_epoch
+
+ lr = get_lr(config)
+
+ opt = Momentum(params=get_param_groups(network), momentum=config.momentum, learning_rate=Tensor(lr),
+ weight_decay=config.weight_decay, loss_scale=config.loss_scale)
+
+ network = TrainingWrapper(network, opt, config.loss_scale // 2)
+ network.set_train()
+
+ if config.rank_save_ckpt_flag:
+ # checkpoint save
+ ckpt_max_num = config.max_epoch * config.steps_per_epoch // config.ckpt_interval
+ ckpt_config = CheckpointConfig(save_checkpoint_steps=config.ckpt_interval, keep_checkpoint_max=1)
+ save_ckpt_path = os.path.join(config.outputs_dir, 'ckpt_' + str(config.rank) + '/')
+ ckpt_cb = ModelCheckpoint(config=ckpt_config, directory=save_ckpt_path, prefix='{}'.format(config.rank))
+ cb_params = _InternalCallbackParam()
+ cb_params.train_network = network
+ cb_params.epoch_num = ckpt_max_num
+ cb_params.cur_epoch_num = 1
+ run_context = RunContext(cb_params)
+ ckpt_cb.begin(run_context)
+
+ old_progress = -1
+ t_end = time.time()
+ data_loader = ds.create_dict_iterator(output_numpy=True, num_epochs=1)
+
+ for i, data in enumerate(data_loader):
+ images = data["image"]
+ input_shape = images.shape[2:4]
+ images = Tensor.from_numpy(images)
+ batch_y_true_0 = Tensor.from_numpy(data['bbox1'])
+ batch_y_true_1 = Tensor.from_numpy(data['bbox2'])
+ batch_y_true_2 = Tensor.from_numpy(data['bbox3'])
+ batch_gt_box0 = Tensor.from_numpy(data['gt_box1'])
+ batch_gt_box1 = Tensor.from_numpy(data['gt_box2'])
+ batch_gt_box2 = Tensor.from_numpy(data['gt_box3'])
+ input_shape = Tensor(tuple(input_shape[::-1]), ms.float32)
+ loss = network(images, batch_y_true_0, batch_y_true_1, batch_y_true_2, batch_gt_box0, batch_gt_box1,
+ batch_gt_box2, input_shape)
+ loss_meter.update(loss.asnumpy())
+
+ if config.rank_save_ckpt_flag:
+ # ckpt progress
+ cb_params.cur_step_num = i + 1 # current step number
+ cb_params.batch_num = i + 2
+ ckpt_cb.step_end(run_context)
+
+ if i % config.log_interval == 0:
+ time_used = time.time() - t_end
+ epoch = int(i / config.steps_per_epoch)
+ fps = config.per_batch_size * (i - old_progress) * config.group_size / time_used
+ if config.rank == 0:
+ config.logger.info('epoch[{}], iter[{}], {}, fps:{:.2f} imgs/sec, '
+ 'lr:{}'.format(epoch, i, loss_meter, fps, lr[i]))
+ t_end = time.time()
+ loss_meter.reset()
+ old_progress = i
+
+ if (i + 1) % config.steps_per_epoch == 0 and config.rank_save_ckpt_flag:
+ cb_params.cur_epoch_num += 1
+
+ if config.is_modelArts:
+ mox.file.copy_parallel(src_url='/cache/outputs/', dst_url='obs://hit-cyf/yolov5_npu/outputs/')
+ config.logger.info('==========end training===============')
+
+if __name__ == "__main__":
+ run_train()