!3728 [东北大学][高校贡献][Mindspore][yolov3_tiny]-迁移训练模型提交

Merge pull request !3728 from 冷情灬/yolov3_tiny

research/cv/yolov3_tiny/README_CN.md

@@ -44,7 +44,8 @@ YOLOv3 Tiny是YOLOv3的一个轻量级变体，它使用池化层并减少卷积
 
 # 数据集
 
-使用的数据集：[COCO 2017](<http://images.cocodataset.org/>)
+使用的数据集：
+[COCO 2017](<http://images.cocodataset.org/>)
 
 - 数据集大小：19 GB
     - 训练集：18 GB，118000张图片
@@ -53,6 +54,13 @@ YOLOv3 Tiny是YOLOv3的一个轻量级变体，它使用池化层并减少卷积
 - 数据格式：图片和json文件
     - 标注：数据在dataset.py中处理。
 
+[face-mask-detection](<https://www.kaggle.com/datasets/andrewmvd/face-mask-detection>)(迁移学习使用)
+
+- 数据集大小: 397.65MB, 853张3类彩色图像
+- 数据格式:RGB图像.
+
+    - 注意：数据将在src/dataset.py 中被处理
+
 - 数据集
 
     1. 目录结构如下：
@@ -431,4 +439,120 @@ YOLOv3-tiny应用于5000张图像上（标注和数据格式必须与COCO val 20
 
 # ModelZoo主页
 
- 请浏览官网[主页](https://gitee.com/mindspore/models)。
+请浏览官网[主页](https://gitee.com/mindspore/models)。
+
+## [迁移学习](#content)
+
+### [迁移学习训练流程](#content)
+
+#### 数据集处理
+
+[数据集下载地址](https://www.kaggle.com/datasets/andrewmvd/face-mask-detection)
+
+下载数据集后解压至yolov3_tiny的dataset目录下，进入src目录，使用data_split脚本划分出80%的训练集和20%的测试集，使用voc脚本生成训练集测试集标签的json文件
+
+```bash
+运行脚本示例
+cd src
+python data_split.py
+python voc.py
+```
+
+```text
+数据集结构
+└─dataset
+  ├─train
+  ├─val
+  ├─annotations
+  ├─images
+  ├─facemask
+  ├─annotations_json
+
+```
+
+```text
+训练前，为yaml文件配置好facemask数据集路径
+# your dataset dir
+dataset_root: /home/mindspore/yolov3_tiny/dataset/
+```
+
+#### 迁移学习训练过程
+
+需要先从[Mindspore Hub](https://www.mindspore.cn/resources/hub/details?MindSpore/1.8/yolov3tiny_coco2017)下载预训练的ckpt
+
+```text
+# 在finetune_config.yaml设置预训练模型的ckpt
+finetune_path: "/home/mindspore/yolov3_tiny/LoadPretrainedModel/yolov3tiny_ascend_v180_coco2017_research_cv_mAP17.5_AP50acc36.0.ckpt""
+```
+
+```bash
+#运行迁移学习训练脚本
+python train.py --config_path  './config/finetune_config.yaml'
+```
+
+**结果展示**
+
+训练结果将存储在示例路径中。checkpoint将存储在 `./outputs/%Y-%m-%d_time_%H_%M_%S/ckpt_0` 路径下，训练loss输出示例如下：
+
+```text
+2022-09-28 00:12:14,528:INFO:epoch[0], iter[0], loss:7826.489258, fps:0.38 imgs/sec, epoch time:84333.98 ms, per step time:84333.98 ms, lr:7.142857043618278e-07
+2022-09-28 00:12:25,968:INFO:epoch[1], iter[21], loss:1231.710783, fps:58.77 imgs/sec, epoch time:11435.29 ms, per step time:544.54 ms, lr:1.5714285837020725e-05
+2022-09-28 00:12:36,381:INFO:epoch[2], iter[42], loss:185.153160, fps:64.54 imgs/sec, epoch time:10411.45 ms, per step time:495.78 ms, lr:3.0714287277078256e-05
+2022-09-28 00:12:46,107:INFO:epoch[3], iter[63], loss:150.027618, fps:69.12 imgs/sec, epoch time:9722.47 ms, per step time:462.97 ms, lr:4.571428507915698e-05
+2022-09-28 00:12:55,684:INFO:epoch[4], iter[84], loss:128.305368, fps:70.22 imgs/sec, epoch time:9569.61 ms, per step time:455.70 ms, lr:6.071428651921451e-05
+2022-09-28 00:13:05,645:INFO:epoch[5], iter[105], loss:123.486393, fps:67.46 imgs/sec, epoch time:9960.92 ms, per step time:474.33 ms, lr:7.571428432129323e-05
+...
+...
+```
+
+#### 迁移学习推理过程
+
+```bash
+#运行迁移学习训练脚本
+python eval.py --config_path  './config/finetune_config.yaml'
+```
+
+**结果展示**
+
+```text
+=============coco eval reulst=========
+  Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.476
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.769
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.560
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.366
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.642
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.588
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.297
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.512
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.548
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.461
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.680
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.601
+
+======================================
+
+mAP: 0.47645057246310263
+
+2022-09-27 23:31:00,800:INFO:testing cost time 0.01h
+
+```
+
+#### 迁移学习quick_start
+
+运行eval脚本后，会生成`predictions.json`文件，在finetune_config.yaml修改`predictions.json`文件的路径后再运行
+predict_path: "/home/mindspore/yolov3_tiny/outputs/%Y-%m-%d_time_%H_%M_%S/predict_%Y-%m-%d_time_%H_%M_%S.json"
+
+```bash
+# 运行quick_start脚本示例
+python quick_start.py --config_path './config/finetune_config.yaml'
+```
+
+**结果说明**
+图中颜色的含义分别是：
+
+- 浅蓝: 真实标签的mask_weared_incorrect
+- 浅绿: 真实标签的with_mask
+- 浅红: 真实标签的without_mask
+- 蓝色: 预测标签的mask_weared_incorrect
+- 绿色: 预测标签的with_mask
+- 红色: 预测标签的without_mask

research/cv/yolov3_tiny/config/finetune_config.yaml

+# 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
+data_path: "/cache/data"
+output_path: "/cache/train"
+load_path: "/cache/checkpoint_path"
+device_target: "CPU" # ['Ascend', 'GPU', 'CPU']
+need_modelarts_dataset_unzip: True
+modelarts_dataset_unzip_name: "coco2017"
+
+# ==============================================================================
+# Training options
+finetune: 1
+# dataset related
+data_dir: "./dataset/"
+finetune_path: "./LoadPretrainedModel/yolov3tiny_ascend_v180_coco2017_research_cv_mAP17.5_AP50acc36.0.ckpt"
+predict_path: "./outputs/2022-09-19_time_03_45_17/predict_2022_09_19_03_45_36.json"
+per_batch_size: 32
+
+# network related
+pretrained_backbone: ""
+resume_yolo: ""
+resume_epoch: 0
+
+# optimizer and lr related
+lr_scheduler: "cosine_annealing"
+lr: 0.0003
+lr_epochs: "220,250"
+lr_gamma: 0.1
+eta_min: 0.0
+t_max: 300
+max_epoch: 300
+warmup_epochs: 20
+weight_decay: 0.016
+momentum: 0.9
+
+# loss related
+loss_scale: 1
+label_smooth: 0
+label_smooth_factor: 0.1
+
+# logging related
+log_interval: 21
+ckpt_path: "./outputs/"
+ckpt_interval: -1
+is_save_on_master: 1
+
+# distributed related
+is_distributed: 0
+rank: 0
+group_size: 1
+
+# profiler init
+need_profiler: 0
+
+# reset default config
+training_shape: 640
+
+
+# Eval option
+pretrained: "./outputs/2022-09-19_time_02_05_24/ckpt_0/yolov3_tiny-300_6300.ckpt"
+log_path: "./outputs/"
+nms_thresh: 0.5
+ann_file: ""
+testing_shape: 640
+eval_ignore_threshold: 0.001
+result_path: ""
+img_path: ""
+
+# Export option
+device_id: 0
+batch_size: 1
+ckpt_file: ""
+file_name: "yolov3_tiny"
+file_format: "MINDIR" # ["AIR", "ONNX", "MINDIR"]
+keep_detect: True
+
+# convert weight option
+input_file: ""
+output_file: ""
+
+# Other default config
+num_parallel_workers: 8
+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]
+              ]
+
+num_classes: 3
+out_channel: 24 #3 * (num_classes + 5)
+max_box: 90
+coord_scale: 4
+large_scale: 4
+
+# confidence under ignore_threshold means no object when training
+ignore_threshold: 0.7
+
+# h->w
+anchor_scales: [[10, 14],
+                [23, 27],
+                [37, 58],
+                [81, 82],
+                [135, 169],
+                [344, 319]]
+
+# test_param
+test_img_shape: [640, 640]
+multi_label: True
+multi_label_thresh: 0.15
+---
+
+# Help description for each configuration
+data_dir: "Train dataset directory."
+per_batch_size: "Batch size for Training."
+pretrained_backbone: "The ckpt file of backbone."
+resume_yolo: "The ckpt file of YOLO, which used to fine tune."
+
+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."
+
+# eval option
+pretrained: "model_path, local pretrained model to load."
+log_path: "checkpoint save location."
+nms_thresh: "threshold for NMS."
+ann_file: "path to annotation."
+testing_shape: "shape for test."
+eval_ignore_threshold: "threshold to throw low quality boxes for eval."
+multi_label: "whether to use multi label."
+multi_label_thresh: "threshold to throw low quality boxes."
+
+# export option
+device_id: "Device id"
+batch_size: "batch size"
+ckpt_file: "Checkpoint file path."
+file_name: "output file name."
+file_format: "file format choices in ['AIR', 'ONNX', 'MINDIR']"
+device_target: "device target. choices in ['Ascend', 'GPU'] for train. choices in ['Ascend', 'GPU', 'CPU'] for export."
+keep_detect: "keep the detect module or not, default: True"
+
+# convert weight option
+input_file: "input file path."
+output_file: "output file path."

research/cv/yolov3_tiny/dataset/facemask/train.txt

+maksssksksss304
+maksssksksss311
+maksssksksss6
+maksssksksss479
+maksssksksss231
+maksssksksss180
+maksssksksss485
+maksssksksss557
+maksssksksss290
+maksssksksss435
+maksssksksss703
+maksssksksss827
+maksssksksss129
+maksssksksss635
+maksssksksss5
+maksssksksss559
+maksssksksss412
+maksssksksss235
+maksssksksss80
+maksssksksss312
+maksssksksss810
+maksssksksss65
+maksssksksss179
+maksssksksss823
+maksssksksss770
+maksssksksss279
+maksssksksss433
+maksssksksss68
+maksssksksss423
+maksssksksss246
+maksssksksss72
+maksssksksss769
+maksssksksss287
+maksssksksss817
+maksssksksss551
+maksssksksss831
+maksssksksss742
+maksssksksss509
+maksssksksss153
+maksssksksss760
+maksssksksss232
+maksssksksss452
+maksssksksss402
+maksssksksss841
+maksssksksss343
+maksssksksss754
+maksssksksss144
+maksssksksss546
+maksssksksss143
+maksssksksss344
+maksssksksss115
+maksssksksss40
+maksssksksss467
+maksssksksss320
+maksssksksss220
+maksssksksss106
+maksssksksss360
+maksssksksss591
+maksssksksss151
+maksssksksss191
+maksssksksss768
+maksssksksss211
+maksssksksss254
+maksssksksss470
+maksssksksss372
+maksssksksss594
+maksssksksss150
+maksssksksss379
+maksssksksss533
+maksssksksss730
+maksssksksss271
+maksssksksss339
+maksssksksss613
+maksssksksss465
+maksssksksss101
+maksssksksss634
+maksssksksss364
+maksssksksss397
+maksssksksss161
+maksssksksss458
+maksssksksss583
+maksssksksss783
+maksssksksss700
+maksssksksss193
+maksssksksss60
+maksssksksss328
+maksssksksss299
+maksssksksss112
+maksssksksss720
+maksssksksss371
+maksssksksss346
+maksssksksss99
+maksssksksss385
+maksssksksss182
+maksssksksss718
+maksssksksss166
+maksssksksss369
+maksssksksss650
+maksssksksss575
+maksssksksss210
+maksssksksss242
+maksssksksss53
+maksssksksss787
+maksssksksss87
+maksssksksss838
+maksssksksss431
+maksssksksss156
+maksssksksss125
+maksssksksss274
+maksssksksss582
+maksssksksss336
+maksssksksss209
+maksssksksss599
+maksssksksss499
+maksssksksss54
+maksssksksss713
+maksssksksss812
+maksssksksss181
+maksssksksss398
+maksssksksss520
+maksssksksss498
+maksssksksss269
+maksssksksss744
+maksssksksss100
+maksssksksss155
+maksssksksss565
+maksssksksss159
+maksssksksss534
+maksssksksss692
+maksssksksss306
+maksssksksss300
+maksssksksss654
+maksssksksss47
+maksssksksss729
+maksssksksss609
+maksssksksss98
+maksssksksss846
+maksssksksss679
+maksssksksss142
+maksssksksss704
+maksssksksss597
+maksssksksss266
+maksssksksss111
+maksssksksss669
+maksssksksss177
+maksssksksss736
+maksssksksss523
+maksssksksss419
+maksssksksss672
+maksssksksss49
+maksssksksss123
+maksssksksss316
+maksssksksss35
+maksssksksss31
+maksssksksss538
+maksssksksss821
+maksssksksss794
+maksssksksss445
+maksssksksss282
+maksssksksss365
+maksssksksss624
+maksssksksss788
+maksssksksss614
+maksssksksss711
+maksssksksss775
+maksssksksss510
+maksssksksss250
+maksssksksss767
+maksssksksss622
+maksssksksss74
+maksssksksss313
+maksssksksss302
+maksssksksss747
+maksssksksss542
+maksssksksss58
+maksssksksss107
+maksssksksss800
+maksssksksss27
+maksssksksss118
+maksssksksss217
+maksssksksss464
+maksssksksss386
+maksssksksss592
+maksssksksss76
+maksssksksss749
+maksssksksss244
+maksssksksss641
+maksssksksss491
+maksssksksss356
+maksssksksss260
+maksssksksss836
+maksssksksss719
+maksssksksss381
+maksssksksss199
+maksssksksss685
+maksssksksss61
+maksssksksss640
+maksssksksss34
+maksssksksss573
+maksssksksss439
+maksssksksss502
+maksssksksss48
+maksssksksss0
+maksssksksss105
+maksssksksss653
+maksssksksss513
+maksssksksss46
+maksssksksss496
+maksssksksss252
+maksssksksss756
+maksssksksss598
+maksssksksss503
+maksssksksss705
+maksssksksss495
+maksssksksss114
+maksssksksss798
+maksssksksss24
+maksssksksss411
+maksssksksss626
+maksssksksss665
+maksssksksss213
+maksssksksss92
+maksssksksss497
+maksssksksss443
+maksssksksss476
+maksssksksss78
+maksssksksss317
+maksssksksss103
+maksssksksss396
+maksssksksss442
+maksssksksss505
+maksssksksss521
+maksssksksss585
+maksssksksss247
+maksssksksss446
+maksssksksss368
+maksssksksss579
+maksssksksss514
+maksssksksss425
+maksssksksss238
+maksssksksss531
+maksssksksss261
+maksssksksss709
+maksssksksss690
+maksssksksss701
+maksssksksss772
+maksssksksss643
+maksssksksss204
+maksssksksss668
+maksssksksss354
+maksssksksss671
+maksssksksss126
+maksssksksss327
+maksssksksss4
+maksssksksss349
+maksssksksss795
+maksssksksss140
+maksssksksss586
+maksssksksss295
+maksssksksss222
+maksssksksss822
+maksssksksss791
+maksssksksss239
+maksssksksss779
+maksssksksss197
+maksssksksss228
+maksssksksss563
+maksssksksss777
+maksssksksss689
+maksssksksss293
+maksssksksss681
+maksssksksss484
+maksssksksss285
+maksssksksss830
+maksssksksss276
+maksssksksss590
+maksssksksss212
+maksssksksss352
+maksssksksss847
+maksssksksss617
+maksssksksss721
+maksssksksss796
+maksssksksss663
+maksssksksss89
+maksssksksss621
+maksssksksss265
+maksssksksss172
+maksssksksss82
+maksssksksss75
+maksssksksss815
+maksssksksss294
+maksssksksss8
+maksssksksss407
+maksssksksss405
+maksssksksss471
+maksssksksss473
+maksssksksss544
+maksssksksss422
+maksssksksss79
+maksssksksss154
+maksssksksss387
+maksssksksss811
+maksssksksss226
+maksssksksss2
+maksssksksss255
+maksssksksss121
+maksssksksss852
+maksssksksss307
+maksssksksss62
+maksssksksss149
+maksssksksss577
+maksssksksss615
+maksssksksss803
+maksssksksss560
+maksssksksss673
+maksssksksss581
+maksssksksss589
+maksssksksss288
+maksssksksss472
+maksssksksss259
+maksssksksss684
+maksssksksss481
+maksssksksss73
+maksssksksss38
+maksssksksss670
+maksssksksss418
+maksssksksss141
+maksssksksss776
+maksssksksss120
+maksssksksss176
+maksssksksss429
+maksssksksss175
+maksssksksss807
+maksssksksss432
+maksssksksss102
+maksssksksss451
+maksssksksss205
+maksssksksss64
+maksssksksss131
+maksssksksss171
+maksssksksss127
+maksssksksss697
+maksssksksss128
+maksssksksss45
+maksssksksss462
+maksssksksss41
+maksssksksss620
+maksssksksss605
+maksssksksss198
+maksssksksss146
+maksssksksss556
+maksssksksss200
+maksssksksss716
+maksssksksss708
+maksssksksss819
+maksssksksss580
+maksssksksss221
+maksssksksss138
+maksssksksss383
+maksssksksss734
+maksssksksss168
+maksssksksss816
+maksssksksss52
+maksssksksss647
+maksssksksss826
+maksssksksss676
+maksssksksss415
+maksssksksss629
+maksssksksss455
+maksssksksss724
+maksssksksss454
+maksssksksss694
+maksssksksss404
+maksssksksss529
+maksssksksss348
+maksssksksss377
+maksssksksss536
+maksssksksss623
+maksssksksss753
+maksssksksss469
+maksssksksss28
+maksssksksss687
+maksssksksss183
+maksssksksss660
+maksssksksss447
+maksssksksss43
+maksssksksss801
+maksssksksss438
+maksssksksss391
+maksssksksss39
+maksssksksss436
+maksssksksss764
+maksssksksss548
+maksssksksss33
+maksssksksss487
+maksssksksss541
+maksssksksss426
+maksssksksss361
+maksssksksss851
+maksssksksss532
+maksssksksss202
+maksssksksss324
+maksssksksss223
+maksssksksss839
+maksssksksss196
+maksssksksss508
+maksssksksss113
+maksssksksss263
+maksssksksss237
+maksssksksss738
+maksssksksss395
+maksssksksss603
+maksssksksss611
+maksssksksss342
+maksssksksss357
+maksssksksss178
+maksssksksss848
+maksssksksss315
+maksssksksss727
+maksssksksss173
+maksssksksss403
+maksssksksss417
+maksssksksss225
+maksssksksss298
+maksssksksss771
+maksssksksss695
+maksssksksss666
+maksssksksss384
+maksssksksss525
+maksssksksss362
+maksssksksss427
+maksssksksss825
+maksssksksss340
+maksssksksss408
+maksssksksss743
+maksssksksss273
+maksssksksss698
+maksssksksss434
+maksssksksss482
+maksssksksss608
+maksssksksss828
+maksssksksss170
+maksssksksss558
+maksssksksss488
+maksssksksss799
+maksssksksss658
+maksssksksss50
+maksssksksss530
+maksssksksss524
+maksssksksss607
+maksssksksss19
+maksssksksss227
+maksssksksss518
+maksssksksss351
+maksssksksss284
+maksssksksss376
+maksssksksss390
+maksssksksss782
+maksssksksss793
+maksssksksss486
+maksssksksss394
+maksssksksss414
+maksssksksss303
+maksssksksss850
+maksssksksss258
+maksssksksss382
+maksssksksss774
+maksssksksss256
+maksssksksss334
+maksssksksss117
+maksssksksss601
+maksssksksss595
+maksssksksss766
+maksssksksss331
+maksssksksss84
+maksssksksss657
+maksssksksss707
+maksssksksss725
+maksssksksss460
+maksssksksss547
+maksssksksss430
+maksssksksss428
+maksssksksss494
+maksssksksss501
+maksssksksss528
+maksssksksss638
+maksssksksss194
+maksssksksss814
+maksssksksss201
+maksssksksss91
+maksssksksss535
+maksssksksss696
+maksssksksss122
+maksssksksss675
+maksssksksss834
+maksssksksss16
+maksssksksss748
+maksssksksss17
+maksssksksss137
+maksssksksss169
+maksssksksss157
+maksssksksss130
+maksssksksss97
+maksssksksss792
+maksssksksss11
+maksssksksss292
+maksssksksss189
+maksssksksss20
+maksssksksss55
+maksssksksss478
+maksssksksss308
+maksssksksss844
+maksssksksss132
+maksssksksss18
+maksssksksss353
+maksssksksss587
+maksssksksss77
+maksssksksss81
+maksssksksss70
+maksssksksss248
+maksssksksss576
+maksssksksss627
+maksssksksss489
+maksssksksss540
+maksssksksss527
+maksssksksss190
+maksssksksss566
+maksssksksss648
+maksssksksss251
+maksssksksss512
+maksssksksss67
+maksssksksss51
+maksssksksss569
+maksssksksss29
+maksssksksss309
+maksssksksss840
+maksssksksss319
+maksssksksss733
+maksssksksss187
+maksssksksss506
+maksssksksss268
+maksssksksss270
+maksssksksss219
+maksssksksss283
+maksssksksss835
+maksssksksss374
+maksssksksss740
+maksssksksss69
+maksssksksss262
+maksssksksss236
+maksssksksss214
+maksssksksss490
+maksssksksss759
+maksssksksss230
+maksssksksss459
+maksssksksss568
+maksssksksss596
+maksssksksss37
+maksssksksss71
+maksssksksss165
+maksssksksss570
+maksssksksss642
+maksssksksss842
+maksssksksss325
+maksssksksss763
+maksssksksss564
+maksssksksss735
+maksssksksss281
+maksssksksss388
+maksssksksss752
+maksssksksss57
+maksssksksss824
+maksssksksss572
+maksssksksss802
+maksssksksss780
+maksssksksss618
+maksssksksss765
+maksssksksss784
+maksssksksss543
+maksssksksss332
+maksssksksss400
+maksssksksss731
+maksssksksss457
+maksssksksss440
+maksssksksss667
+maksssksksss550
+maksssksksss7
+maksssksksss88
+maksssksksss683
+maksssksksss366
+maksssksksss655
+maksssksksss682
+maksssksksss466
+maksssksksss21
+maksssksksss240
+maksssksksss755
+maksssksksss483
+maksssksksss843
+maksssksksss578
+maksssksksss184
+maksssksksss275
+maksssksksss833
+maksssksksss804
+maksssksksss399
+maksssksksss820
+maksssksksss593
+maksssksksss162
+maksssksksss549
+maksssksksss410
+maksssksksss691
+maksssksksss253
+maksssksksss66
+maksssksksss715
+maksssksksss124
+maksssksksss604
+maksssksksss659
+maksssksksss164
+maksssksksss633
+maksssksksss305
+maksssksksss739
+maksssksksss552
+maksssksksss109
+maksssksksss152
+maksssksksss245
+maksssksksss448
+maksssksksss94
+maksssksksss337
+maksssksksss688
+maksssksksss728
+maksssksksss1
+maksssksksss453
+maksssksksss504
+maksssksksss335
+maksssksksss389
+maksssksksss286
+maksssksksss93
+maksssksksss367
+maksssksksss741
+maksssksksss188
+maksssksksss463
+maksssksksss500
+maksssksksss370
+maksssksksss373
+maksssksksss380
+maksssksksss218
+maksssksksss686
+maksssksksss750
+maksssksksss321
+maksssksksss516
+maksssksksss322
+maksssksksss761
+maksssksksss257
+maksssksksss63
+maksssksksss845
+maksssksksss359
+maksssksksss519
+maksssksksss23
+maksssksksss229
+maksssksksss714
+maksssksksss562
+maksssksksss10
+maksssksksss233
+maksssksksss409
+maksssksksss264
+maksssksksss329
+maksssksksss289
+maksssksksss646
+maksssksksss36
+maksssksksss797
+maksssksksss330
+maksssksksss849
+maksssksksss318
+maksssksksss805
+maksssksksss139
+maksssksksss545
+maksssksksss134
+maksssksksss474
+maksssksksss174
+maksssksksss86
+maksssksksss280
+maksssksksss644
+maksssksksss449

research/cv/yolov3_tiny/dataset/facemask/val.txt

+maksssksksss461
+maksssksksss314
+maksssksksss456
+maksssksksss203
+maksssksksss185
+maksssksksss751
+maksssksksss85
+maksssksksss758
+maksssksksss680
+maksssksksss652
+maksssksksss249
+maksssksksss781
+maksssksksss167
+maksssksksss416
+maksssksksss539
+maksssksksss722
+maksssksksss808
+maksssksksss600
+maksssksksss207
+maksssksksss135
+maksssksksss25
+maksssksksss674
+maksssksksss610
+maksssksksss406
+maksssksksss186
+maksssksksss441
+maksssksksss413
+maksssksksss277
+maksssksksss710
+maksssksksss363
+maksssksksss444
+maksssksksss116
+maksssksksss12
+maksssksksss350
+maksssksksss420
+maksssksksss136
+maksssksksss30
+maksssksksss706
+maksssksksss631
+maksssksksss517
+maksssksksss637
+maksssksksss636
+maksssksksss818
+maksssksksss699
+maksssksksss378
+maksssksksss693
+maksssksksss492
+maksssksksss661
+maksssksksss192
+maksssksksss14
+maksssksksss345
+maksssksksss625
+maksssksksss567
+maksssksksss145
+maksssksksss243
+maksssksksss301
+maksssksksss341
+maksssksksss790
+maksssksksss326
+maksssksksss278
+maksssksksss468
+maksssksksss522
+maksssksksss216
+maksssksksss32
+maksssksksss726
+maksssksksss13
+maksssksksss555
+maksssksksss160
+maksssksksss477
+maksssksksss554
+maksssksksss56
+maksssksksss746
+maksssksksss493
+maksssksksss310
+maksssksksss809
+maksssksksss26
+maksssksksss553
+maksssksksss612
+maksssksksss110
+maksssksksss9
+maksssksksss475
+maksssksksss737
+maksssksksss338
+maksssksksss662
+maksssksksss829
+maksssksksss272
+maksssksksss215
+maksssksksss450
+maksssksksss375
+maksssksksss574
+maksssksksss507
+maksssksksss323
+maksssksksss702
+maksssksksss392
+maksssksksss732
+maksssksksss537
+maksssksksss333
+maksssksksss108
+maksssksksss561
+maksssksksss678
+maksssksksss757
+maksssksksss104
+maksssksksss515
+maksssksksss639
+maksssksksss347
+maksssksksss158
+maksssksksss355
+maksssksksss588
+maksssksksss806
+maksssksksss206
+maksssksksss393
+maksssksksss83
+maksssksksss571
+maksssksksss22
+maksssksksss42
+maksssksksss3
+maksssksksss649
+maksssksksss44
+maksssksksss656
+maksssksksss480
+maksssksksss224
+maksssksksss526
+maksssksksss778
+maksssksksss723
+maksssksksss119
+maksssksksss195
+maksssksksss712
+maksssksksss133
+maksssksksss297
+maksssksksss786
+maksssksksss762
+maksssksksss437
+maksssksksss745
+maksssksksss606
+maksssksksss59
+maksssksksss241
+maksssksksss616
+maksssksksss664
+maksssksksss15
+maksssksksss584
+maksssksksss773
+maksssksksss96
+maksssksksss163
+maksssksksss837
+maksssksksss421
+maksssksksss511
+maksssksksss291
+maksssksksss717
+maksssksksss785
+maksssksksss632
+maksssksksss630
+maksssksksss234
+maksssksksss832
+maksssksksss677
+maksssksksss602
+maksssksksss147
+maksssksksss358
+maksssksksss401
+maksssksksss628
+maksssksksss424
+maksssksksss789
+maksssksksss95
+maksssksksss208
+maksssksksss645
+maksssksksss148
+maksssksksss90
+maksssksksss651
+maksssksksss813
+maksssksksss619
+maksssksksss267

research/cv/yolov3_tiny/eval.py

@@ -20,12 +20,11 @@ import time
 from collections import defaultdict
 
 import numpy as np
+import mindspore as ms
 from mindspore import Tensor
 from mindspore import context
 from mindspore import dtype as mstype
 from mindspore.context import ParallelMode
-from mindspore.train.serialization import load_checkpoint
-from mindspore.train.serialization import load_param_into_net
 from pycocotools.coco import COCO
 from pycocotools.cocoeval import COCOeval
 
@@ -57,16 +56,19 @@ class DetectionEngine:
 
     def __init__(self, config_detection):
         self.eval_ignore_threshold = config_detection.eval_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']
+        if config.finetune:
+            self.labels = ['without_mask', 'with_mask', 'mask_weared_incorrect']
+        else:
+            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.num_classes = len(self.labels)
         self.results = {}
         self.file_path = ''
@@ -193,6 +195,9 @@ class DetectionEngine:
         stdout = sys.stdout
         sys.stdout = rdct
         coco_eval.summarize()
+        mAP = coco_eval.stats[0]
+        print("\n======================================\n")
+        print(f"mAP: {mAP}")
         sys.stdout = stdout
         return rdct.content
 
@@ -341,8 +346,12 @@ def modelarts_pre_process():
 def run_test():
     """The function of eval."""
     start_time = time.time()
-    config.data_root = os.path.join(config.data_dir, 'val2017')
-    config.ann_file = os.path.join(config.data_dir, 'annotations/instances_val2017.json')
+    if config.finetune:
+        config.data_root = os.path.join(config.data_dir, 'val/images')
+        config.ann_file = os.path.join(config.data_dir, 'annotations_json/val.json')
+    else:
+        config.data_root = os.path.join(config.data_dir, 'val2017')
+        config.ann_file = os.path.join(config.data_dir, 'annotations/instances_val2017.json')
 
     device_id = int(os.getenv('DEVICE_ID')) if os.getenv('DEVICE_ID') else 0
     # device_id = 1
@@ -363,7 +372,7 @@ def run_test():
 
     config.logger.info(config.pretrained)
     if os.path.isfile(config.pretrained):
-        param_dict = load_checkpoint(config.pretrained)
+        param_dict = ms.load_checkpoint(config.pretrained)
         param_dict_new = {}
         for key, values in param_dict.items():
             if key.startswith('moments.'):
@@ -372,7 +381,7 @@ def run_test():
                 param_dict_new[key[13:]] = values
             else:
                 param_dict_new[key] = values
-        load_param_into_net(network, param_dict_new)
+        ms.load_param_into_net(network, param_dict_new)
         config.logger.info('load_model %s success', config.pretrained)
     else:
         config.logger.info('%s not exists or not a pre-trained file', config.pretrained)

research/cv/yolov3_tiny/quick_start.py

+# Copyright 2022 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.
+# ============================================================================
+
+"""visualize for yolov3_tiny"""
+
+import random
+import cv2
+import matplotlib.pyplot as plt
+from pycocotools.coco import COCO
+from model_utils.config import config
+
+random.seed(11)
+pred_blue = (0, 0, 255)  # 'with_mask': 1
+true_blue = (0, 191, 255)
+pred_red = (255, 0, 0)  # 'without_mask': 2,
+true_red = (255, 140, 0)
+pred_green = (0, 255, 0)  # 'mask_weared_incorrect': 3
+true_green = (0, 128, 0)
+
+
+def visualize_model():
+    # load best ckpt to generate instances_val.json and predictions.json
+
+    dataset_dir = r'./dataset/val/images/'
+    ann_file = './dataset/annotations_json/val.json'
+    coco = COCO(ann_file)
+    catids = coco.getCatIds()
+    imgids = coco.getImgIds()
+    img_list = random.sample(imgids, 8)
+    coco_res = coco.loadRes(config.predict_path)
+    catids_res = coco_res.getCatIds()
+    for i in img_list:
+        img = coco.loadImgs(i)[0]
+        image = cv2.imread(dataset_dir + img['file_name'])
+        image_res = image
+        annids = coco.getAnnIds(imgIds=img['id'], catIds=catids, iscrowd=None)
+        annos = coco.loadAnns(annids)
+        annids_res = coco_res.getAnnIds(imgIds=img['id'], catIds=catids_res, iscrowd=None)
+        annos_res = coco_res.loadAnns(annids_res)
+        plt.figure(figsize=(6, 6))
+        for anno in annos:
+            bbox = anno['bbox']
+            x, y, w, h = bbox
+            if anno['category_id'] == 1:  # with_mask
+                anno_image = cv2.rectangle(image, (int(x), int(y)), (int(x + w), int(y + h)), true_blue, 2)
+            elif anno['category_id'] == 2:  # without_mask
+                anno_image = cv2.rectangle(image, (int(x), int(y)), (int(x + w), int(y + h)), true_red, 2)
+            else:  # mask_weared_incorrect
+                anno_image = cv2.rectangle(image, (int(x), int(y)), (int(x + w), int(y + h)), true_green, 2)
+            plt.subplot(1, 2, 1)
+            plt.plot([-2, 3], [1, 5])
+            plt.title('true')
+            plt.imshow(anno_image)
+        for anno_res in annos_res:
+            bbox_res = anno_res['bbox']
+            x, y, w, h = bbox_res
+            if anno_res['category_id'] == 1:
+                res_image = cv2.rectangle(image_res, (int(x), int(y)), (int(x + w), int(y + h)), pred_blue, 2)
+            elif anno_res['category_id'] == 2:
+                res_image = cv2.rectangle(image_res, (int(x), int(y)), (int(x + w), int(y + h)), pred_red, 2)
+            else:
+                res_image = cv2.rectangle(image_res, (int(x), int(y)), (int(x + w), int(y + h)), pred_green, 2)
+            plt.subplot(1, 2, 2)
+            plt.title('pred')
+            plt.imshow(res_image)
+    plt.show()
+
+
+if __name__ == '__main__':
+    visualize_model()

research/cv/yolov3_tiny/src/data_split.py

+# Copyright 2022 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.
+# ============================================================================
+
+"""data_split"""
+
+import os
+import shutil
+
+image_original_path = '../dataset/images/'
+label_original_path = '../dataset/annotations/'
+
+train_image_path = '../dataset/train/images/'
+train_label_path = '../dataset/train/annotations/'
+
+val_image_path = '../dataset/val/images/'
+val_label_path = '../dataset/val/annotations/'
+
+
+def mkdir():
+    if not os.path.exists(train_image_path):
+        os.makedirs(train_image_path)
+    if not os.path.exists(train_label_path):
+        os.makedirs(train_label_path)
+
+    if not os.path.exists(val_image_path):
+        os.makedirs(val_image_path)
+    if not os.path.exists(val_label_path):
+        os.makedirs(val_label_path)
+
+
+def main():
+    mkdir()
+    with open("../dataset/facemask/train.txt", 'r') as f:
+        for line in f:
+            dst_train_image = train_image_path + line[:-1] + '.jpg'
+            dst_train_label = train_label_path + line[:-1] + '.xml'
+            shutil.copyfile(image_original_path + line[:-1] + '.png', dst_train_image)
+            shutil.copyfile(label_original_path + line[:-1] + '.xml', dst_train_label)
+
+    with open("../dataset/facemask/val.txt", 'r') as f:
+        for line in f:
+            dst_val_image = val_image_path + line[:-1] + '.jpg'
+            dst_val_label = val_label_path + line[:-1] + '.xml'
+            shutil.copyfile(image_original_path + line[:-1] + '.png', dst_val_image)
+            shutil.copyfile(label_original_path + line[:-1] + ".xml", dst_val_label)
+
+
+if __name__ == '__main__':
+    main()

research/cv/yolov3_tiny/src/voc.py

+# Copyright 2022 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.
+# ============================================================================
+
+"""xml covert to coco"""
+
+import glob
+import json
+import os
+import xml.etree.ElementTree as ET
+import numpy as np
+
+define_categories = {'with_mask': 1, 'without_mask': 2, 'mask_weared_incorrect': 3}
+anno_train_dir = "../dataset/train/annotations"
+anno_val_dir = "../dataset/val/annotations"
+save_dir = "../dataset/annotations_json"
+if not os.path.exists(save_dir):
+    os.makedirs(save_dir)
+
+
+def convert(xml_list, json_file):
+    json_dict = {"images": [], "type": "instances", "annotations": [], "categories": []}
+    categories = define_categories.copy()
+    bnd_id = 1
+
+    for line in xml_list:
+        xml_f = line
+        tree = ET.parse(xml_f)
+        root = tree.getroot()
+
+        filename = root.find('filename').text[:-4] + '.jpg'
+
+        image_id = int(filename[12:-4])  # maksssksksss 12
+        size = root.find('size')
+        width = int(size.find('width').text)
+        height = int(size.find('height').text)
+        image = {'file_name': filename, 'height': height, 'width': width, 'id': image_id}
+        json_dict['images'].append(image)
+
+        for obj in root.findall('object'):
+            category = obj.find('name').text
+            if category not in categories:
+                continue
+            category_id = categories[category]
+            bndbox = obj.find('bndbox')
+            xmin = int(float(bndbox.find('xmin').text)) - 1
+            ymin = int(float(bndbox.find('ymin').text)) - 1
+            xmax = int(float(bndbox.find('xmax').text)) - 1
+            ymax = int(float(bndbox.find('ymax').text)) - 1
+            assert (xmax > xmin), "xmax <= xmin, {}".format(line)
+            assert (ymax > ymin), "ymax <= ymin, {}".format(line)
+            o_width = abs(xmax - xmin)
+            o_height = abs(ymax - ymin)
+            ann = {'area': o_width * o_height,
+                   'iscrowd': 0,
+                   'image_id': image_id,
+                   'bbox': [xmin, ymin, o_width, o_height],
+                   'category_id': category_id,
+                   'id': bnd_id,
+                   'segmentation': []}  # Currently we do not support segmentation
+            json_dict['annotations'].append(ann)
+            bnd_id = bnd_id + 1
+
+    for cate_name, cid in categories.items():
+        cat = {'supercategory': 'none',
+               'id': cid,
+               'name': cate_name}
+        json_dict['categories'].append(cat)
+    json_fp = open(json_file, 'w')
+    json_str = json.dumps(json_dict)
+    json_fp.write(json_str)
+    json_fp.close()
+    print("------------create {} done--------------".format(json_file))
+    print("category: id --> {}".format(categories))
+    print(categories.keys())
+    print(categories.values())
+
+
+if __name__ == '__main__':
+    anno_train_list = glob.glob(anno_train_dir + "/*.xml")
+    anno_train_list = np.sort(anno_train_list)
+
+    anno_val_list = glob.glob(anno_val_dir + "/*.xml")
+    anno_val_list = np.sort(anno_val_list)
+
+    # save json files
+    save_anno_train = save_dir + "/train.json"
+    save_anno_val = save_dir + "/val.json"
+
+    convert(anno_train_list, save_anno_train)
+    convert(anno_val_list, save_anno_val)

research/cv/yolov3_tiny/src/yolo.py

@@ -47,8 +47,8 @@ class YOLOv3Tiny(nn.Cell):
         super(YOLOv3Tiny, self).__init__()
 
         self.backbone = backbone
-        self.head1 = nn.Conv2d(512, 255, kernel_size=1, stride=1, has_bias=True)
-        self.head2 = nn.Conv2d(256, 255, kernel_size=1, stride=1, has_bias=True)
+        self.head1 = nn.Conv2d(512, config.out_channel, kernel_size=1, stride=1, has_bias=True)
+        self.head2 = nn.Conv2d(256, config.out_channel, kernel_size=1, stride=1, has_bias=True)
         self.concat = P.Concat(axis=1)
 
     def construct(self, x):

research/cv/yolov3_tiny/train.py

@@ -55,11 +55,12 @@ def set_default():
         config.t_max = config.max_epoch
 
     config.lr_epochs = list(map(int, config.lr_epochs.split(',')))
-    config.data_root = os.path.join(config.data_dir, 'train2017')
-    config.ann_file = os.path.join(config.data_dir, 'annotations/instances_train2017.json')
-
-    config.data_val_root = os.path.join(config.data_dir, 'val2017')
-    config.ann_val_file = os.path.join(config.data_dir, 'annotations/instances_val2017.json')
+    if config.finetune:
+        config.data_root = os.path.join(config.data_dir, 'train/images')
+        config.ann_file = os.path.join(config.data_dir, 'annotations_json/train.json')
+    else:
+        config.data_root = os.path.join(config.data_dir, 'train2017')
+        config.ann_file = os.path.join(config.data_dir, 'annotations/instances_train2017.json')
 
     context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target, device_id=get_device_id())
 
@@ -183,7 +184,28 @@ def prepare_network():
     network = YOLOv3(is_training=True)
     # default is kaiming-normal
     default_recurisive_init(network)
-    load_yolo_params(config, network)
+    if config.finetune:
+        param_dict = ms.load_checkpoint(config.finetune_path)
+        param_dict_new = {}
+        for key, values in param_dict.items():
+            if key.startswith('moments.'):
+                continue
+            elif key.startswith('yolo_network.'):
+                param_dict_new[key[13:]] = values
+            else:
+                param_dict_new[key] = values
+        for key in list(param_dict_new.keys()):
+            if 'feature_map.head1.weight' in key:
+                del param_dict_new[key]
+            if 'feature_map.head1.bias' in key:
+                del param_dict_new[key]
+            if 'feature_map.head2.weight' in key:
+                del param_dict_new[key]
+            if 'feature_map.head2.bias' in key:
+                del param_dict_new[key]
+        ms.load_param_into_net(network, param_dict_new)
+    else:
+        load_yolo_params(config, network)
 
     network = YOLOWithLossCell(network)
     return network
@@ -235,10 +257,10 @@ def run_train():
 
     if config.rank_save_ckpt_flag:
         # checkpoint save
-        ckpt_max_num = config.max_epoch * config.steps_per_epoch // config.ckpt_interval
+        ckpt_max_num = 5 * config.steps_per_epoch // config.ckpt_interval
         ckpt_config = CheckpointConfig(save_checkpoint_steps=config.ckpt_interval, keep_checkpoint_max=ckpt_max_num)
         save_ckpt_path = os.path.join(config.outputs_dir, f'ckpt_{config.rank}/')
-        ckpt_cb = ModelCheckpoint(config=ckpt_config, directory=save_ckpt_path, prefix='{}'.format(config.rank))
+        ckpt_cb = ModelCheckpoint(config=ckpt_config, directory=save_ckpt_path, prefix='yolov3_tiny')
         cb_params = _InternalCallbackParam()
         cb_params.train_network = network
         cb_params.epoch_num = ckpt_max_num