{"id":18527905,"url":"https://github.com/nemonameless/mindsporeyolov5","last_synced_at":"2025-05-14T21:33:19.409Z","repository":{"id":118412571,"uuid":"459130830","full_name":"nemonameless/mindsporeyolov5","owner":"nemonameless","description":null,"archived":false,"fork":false,"pushed_at":"2022-02-14T11:19:12.000Z","size":78,"stargazers_count":1,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"main","last_synced_at":"2025-02-17T06:30:26.760Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/nemonameless.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-02-14T11:16:54.000Z","updated_at":"2022-07-13T12:35:42.000Z","dependencies_parsed_at":null,"dependency_job_id":"abf87268-a480-45ec-b045-457984c8ab33","html_url":"https://github.com/nemonameless/mindsporeyolov5","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/nemonameless%2Fmindsporeyolov5","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/nemonameless%2Fmindsporeyolov5/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/nemonameless%2Fmindsporeyolov5/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/nemonameless%2Fmindsporeyolov5/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/nemonameless","download_url":"https://codeload.github.com/nemonameless/mindsporeyolov5/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":254231406,"owners_count":22036363,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-11-06T17:56:16.544Z","updated_at":"2025-05-14T21:33:19.386Z","avatar_url":"https://github.com/nemonameless.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Inference ProcessContents\n\n- [YOLOv5 Description](#YOLOv5-description)\n- [Model Architecture](#model-architecture)\n- [Dataset](#dataset)\n- [Quick Start](#quick-start)\n- [Script Description](#script-description)\n    - [Script and Sample Code](#script-and-sample-code)\n    - [Script Parameters](#script-parameters)\n    - [Training Process](#training-process)\n        - [Training](#training)\n        - [Distributed Training](#distributed-training)  \n    - [Evaluation Process](#evaluation-process)\n        - [Evaluation](#evaluation)\n    - [Inference Process](#inference-process)\n        - [Export MindIR](#export-mindir)\n        - [Infer on Ascend310](#infer-on-ascend310)\n        - [result](#result)\n- [Model Description](#model-description)\n- [Performance](#performance)  \n    - [Evaluation Performance](#evaluation-performance)\n    - [Inference Performance](#inference-performance)\n    - [Transfer Learning](#transfer-learning)\n- [Description of Random Situation](#description-of-random-situation)\n- [ModelZoo Homepage](#modelzoo-homepage)\n\n# [YOLOv5 Description](#contents)\n\nPublished in April 2020, YOLOv5 achieved state of the art performance on the COCO dataset for object detection. It is an important improvement of YoloV3, the implementation of a new architecture in the **Backbone** and the modifications in the **Neck** have improved the **mAP**(mean Average Precision) by **10%** and the number of **FPS**(Frame per Second) by **12%**.\n\n[code](https://github.com/ultralytics/yolov5)\n\n# [Model Architecture](#contents)\n\nThe YOLOv5 network is mainly composed of CSP and Focus as a backbone, spatial pyramid pooling(SPP) additional module, PANet path-aggregation neck and YOLOv3 head. [CSP](https://arxiv.org/abs/1911.11929) is a novel backbone that can enhance the learning capability of CNN. The [spatial pyramid pooling](https://arxiv.org/abs/1406.4729) block is added over CSP to increase the receptive field and separate out the most significant context features. Instead of Feature pyramid networks (FPN) for object detection used in YOLOv3, the PANet is used as the method for parameter aggregation for different detector levels. To be more specifical, CSPDarknet53 contains 5 CSP modules which use the convolution **C** with kernel size k=3x3, stride s = 2x2; Within the PANet and SPP, **1x1, 5x5, 9x9, 13x13 max poolings are applied.\n\n# [Dataset](#contents)\n\nDataset used: [COCO2017](\u003chttps://cocodataset.org/#download\u003e)\n\nNote that you can run the scripts with **COCO2017 **or any other datasets with the same format as MS COCO Annotation. But we do suggest user to use MS COCO dataset to experience our model.\n\n# [Quick Start](#contents)\n\nAfter installing MindSpore via the official website, you can start training and evaluation as follows:\n\n```bash\n#run training example(1p) on Ascend by python command\npython train.py \\\n    --data_dir=xxx/dataset \\\n    --is_distributed=0 \\\n    --yolov5_version='yolov5s' \\\n    --lr=0.02 \\\n    --max_epoch=300 \\\n    --warmup_epochs=20 \\\n    --per_batch_size=128 \\\n    --lr_scheduler=cosine_annealing \u003e log.txt 2\u003e\u00261 \u0026\n```\n\n```bash\n# For Ascend device, distributed training example(8p) by shell script\nbash run_distribute_train.sh xxx/dateset/ xxx/cspdarknet.ckpt rank_table_8pcs.json\n\n# For GPU device, distributed training example(8p) by shell script\nbash run_distribute_train_gpu.sh xxx/dateset [RANK_SIZE]\n```\n\n```bash\n# run evaluation on Ascend by python command\npython eval.py \\\n    --data_dir=xxx/dataset \\\n    --eval_shape=640 \u003e log.txt 2\u003e\u00261 \u0026\n\n# run evaluation on GPU by python command\npython eval.py \\\n    --device_target=\"GPU\" \\\n    --data_dir=xxx/dataset \\\n    --yolov5_version='yolov5s' \\\n    --pretrained=\"***/*.ckpt\" \\\n    --eval_shape=640 \u003e log.txt 2\u003e\u00261 \u0026\n```\n\n```bash\n# run evaluation on Ascend by shell script\nbash run_eval.sh xxx/dataset xxx/yolov5.ckpt\n```\n\n# [Script Description](#contents)\n\n## [Script and Sample Code](#contents)\n\n```bash\n├── model_zoo\n    ├── README.md                              // descriptions about all the models\n    ├── yolov5\n        ├── README.md                          // descriptions about yolov5\n        ├── scripts\n        │   ├──run_distribute_train.sh         // launch distributed training(8p) in ascend\n        │   ├──run_distribute_train_gpu.sh     // launch distributed training(8p) in GPU\n        │   ├──run_standalone_train.sh         // launch 1p training in ascend\n        │   ├──run_eval.sh                     // shell script for evaluation\n        │   ├──rank_table_8pcs.json            // the example of rank table settings for 8p training\n        ├──model_utils\n        │   ├──config.py                       // getting config parameters\n        │   ├──device_adapter.py               // getting device info\n        │   ├──local_adapter.py                // getting device info\n        │   ├──moxing_adapter.py               // Decorator\n        ├── src\n        │   ├──backbone.py                     // backbone of network\n        │   ├──distributed_sampler.py          // iterator of dataset\n        │   ├──initializer.py                  // initializer of parameters\n        │   ├──logger.py                       // log function\n        │   ├──loss.py                         // loss function\n        │   ├──lr_scheduler.py                 // generate learning rate\n        │   ├──transforms.py                   // Preprocess data\n        │   ├──util.py                         // util function\n        │   ├──yolo.py                         // yolov5 network\n        │   ├──yolo_dataset.py                 // create dataset for YOLOV5\n        ├── default_config.yaml                // parameter configuration\n        ├── train.py                           // training script\n        ├── eval.py                            // evaluation script\n        ├── export.py                          // export script\n```\n\n## [Script Parameters](#contents)\n\n```python\nMajor parameters in train.py are:\n\noptional arguments:\n\n  --device_target       device where the code will be implemented: \"Ascend\", default is \"Ascend\"\n  --data_dir            Train dataset directory.\n  --per_batch_size      Batch size for Training. Default: 8.\n  --pretrained_backbone The ckpt file of CSPDarknet53. Default: \"\".\n  --resume_yolov5       The ckpt file of YOLOv5, which used to fine tune.Default: \"\"\n  --lr_scheduler        Learning rate scheduler, options: exponential,cosine_annealing.\n                        Default: cosine_annealing\n  --lr                  Learning rate. Default: 0.02\n  --lr_epochs           Epoch of changing of lr changing, split with \",\". Default: '220,250'\n  --lr_gamma            Decrease lr by a factor of exponential lr_scheduler. Default: 0.1\n  --eta_min             Eta_min in cosine_annealing scheduler. Default: 0.\n  --t_max               T-max in cosine_annealing scheduler. Default: 320\n  --max_epoch           Max epoch num to train the model. Default: 320\n  --warmup_epochs       Warmup epochs. Default: 20\n  --weight_decay        Weight decay factor. Default: 0.0005\n  --momentum            Momentum. Default: 0.9\n  --loss_scale          Static loss scale. Default: 64\n  --label_smooth        Whether to use label smooth in CE. Default:0\n  --label_smooth_factor Smooth strength of original one-hot. Default: 0.1\n  --log_interval        Logging interval steps. Default: 100\n  --ckpt_path           Checkpoint save location. Default: outputs/\n  --is_distributed      Distribute train or not, 1 for yes, 0 for no. Default: 1\n  --rank                Local rank of distributed. Default: 0\n  --group_size          World size of device. Default: 1\n  --need_profiler       Whether use profiler. 0 for no, 1 for yes. Default: 0\n  --training_shape      Fix training shape. Default: \"\"\n  --resize_rate         Resize rate for multi-scale training. Default: 10\n  --bind_cpu            Whether bind cpu when distributed training. Default: True\n  --device_num          Device numbers per server. Default: 8\n```\n\n## [Training Process](#contents)\n\n### Training\n\nFor Ascend device, standalone training can be started like this:\n\n```python\n#run training example(1p) by python command\npython train.py \\\n    --data_dir=xxx/dataset \\\n    --yolov5_version='yolov5s' \\\n    --is_distributed=0 \\\n    --lr=0.02 \\\n    --max_epoch=300 \\\n    --warmup_epochs=20 \\\n    --per_batch_size=128 \\\n    --lr_scheduler=cosine_annealing \u003e log.txt 2\u003e\u00261 \u0026\n```\n\nYou should fine tune the params when run training 1p on GPU\n\nThe python command above will run in the background, you can view the results through the file `log.txt`.\n\nAfter training, you'll get some checkpoint files under the **outputs** folder by default. The loss value will be achieved as follows:\n\n```python\n# grep \"loss:\" log.txt\n2021-08-06 15:30:15,798:INFO:epoch[0], iter[600], loss:296.308071, fps:44.44 imgs/sec, lr:0.00010661844862625003\n2021-08-06 15:31:21,119:INFO:epoch[0], iter[700], loss:276.071959, fps:48.99 imgs/sec, lr:0.00012435863027349114\n2021-08-06 15:32:26,185:INFO:epoch[0], iter[800], loss:266.955208, fps:49.18 imgs/sec, lr:0.00014209879736881703\n2021-08-06 15:33:30,507:INFO:epoch[0], iter[900], loss:252.610914, fps:49.75 imgs/sec, lr:0.00015983897901605815\n2021-08-06 15:34:42,176:INFO:epoch[0], iter[1000], loss:243.106683, fps:44.65 imgs/sec, lr:0.00017757914611138403\n2021-08-06 15:35:47,429:INFO:epoch[0], iter[1100], loss:240.498834, fps:49.04 imgs/sec, lr:0.00019531932775862515\n2021-08-06 15:36:48,945:INFO:epoch[0], iter[1200], loss:245.711473, fps:52.02 imgs/sec, lr:0.00021305949485395104\n2021-08-06 15:37:51,293:INFO:epoch[0], iter[1300], loss:231.388255, fps:51.33 imgs/sec, lr:0.00023079967650119215\n2021-08-06 15:38:55,680:INFO:epoch[0], iter[1400], loss:238.904242, fps:49.70 imgs/sec, lr:0.00024853984359651804\n2021-08-06 15:39:57,419:INFO:epoch[0], iter[1500], loss:232.161600, fps:51.83 imgs/sec, lr:0.00026628002524375916\n2021-08-06 15:41:03,808:INFO:epoch[0], iter[1600], loss:227.844698, fps:48.20 imgs/sec, lr:0.00028402020689100027\n2021-08-06 15:42:06,155:INFO:epoch[0], iter[1700], loss:226.668858, fps:51.33 imgs/sec, lr:0.00030176035943441093\n...\n```\n\n### Distributed Training\n\nFor Ascend device, distributed training example(8p) by shell script：\n\n```bash\n# For Ascend device, distributed training example(8p) by shell script\nbash run_distribute_train.sh xxx/dateset/ xxx/cspdarknet.ckpt rank_table_8pcs.json\n```\n\nThe above shell script will run distribute training in the background. You can view the results through the file train_parallel[X]/log.txt. The loss value will be achieved as follows:\n\n```bash\n# distribute training result(8p, dynamic shape)\n...\n2021-08-05 16:01:34,116:INFO:epoch[0], iter[200], loss:415.453676, fps:580.07 imgs/sec, lr:0.0002742903889156878\n2021-08-05 16:01:57,588:INFO:epoch[0], iter[300], loss:273.358383, fps:545.96 imgs/sec, lr:0.00041075327317230403\n2021-08-05 16:02:26,247:INFO:epoch[0], iter[400], loss:244.621502, fps:446.64 imgs/sec, lr:0.0005472161574289203\n2021-08-05 16:02:55,532:INFO:epoch[0], iter[500], loss:234.524876, fps:437.10 imgs/sec, lr:0.000683679012581706\n2021-08-05 16:03:25,046:INFO:epoch[0], iter[600], loss:235.185213, fps:434.08 imgs/sec, lr:0.0008201419259421527\n2021-08-05 16:03:54,585:INFO:epoch[0], iter[700], loss:228.878598, fps:433.48 imgs/sec, lr:0.0009566047810949385\n2021-08-05 16:04:23,932:INFO:epoch[0], iter[800], loss:219.259134, fps:436.29 imgs/sec, lr:0.0010930676944553852\n2021-08-05 16:04:52,707:INFO:epoch[0], iter[900], loss:225.741833, fps:444.84 imgs/sec, lr:0.001229530549608171\n2021-08-05 16:05:21,872:INFO:epoch[1], iter[1000], loss:218.811336, fps:438.91 imgs/sec, lr:0.0013659934047609568\n2021-08-05 16:05:51,216:INFO:epoch[1], iter[1100], loss:219.491889, fps:436.50 imgs/sec, lr:0.0015024563763290644\n2021-08-05 16:06:20,546:INFO:epoch[1], iter[1200], loss:219.895906, fps:436.57 imgs/sec, lr:0.0016389192314818501\n2021-08-05 16:06:49,521:INFO:epoch[1], iter[1300], loss:218.516680, fps:441.79 imgs/sec, lr:0.001775382086634636\n2021-08-05 16:07:18,303:INFO:epoch[1], iter[1400], loss:209.922935, fps:444.79 imgs/sec, lr:0.0019118449417874217\n2021-08-05 16:07:47,702:INFO:epoch[1], iter[1500], loss:210.997816, fps:435.60 imgs/sec, lr:0.0020483077969402075\n2021-08-05 16:08:16,482:INFO:epoch[1], iter[1600], loss:210.678421, fps:444.88 imgs/sec, lr:0.002184770768508315\n2021-08-05 16:08:45,568:INFO:epoch[1], iter[1700], loss:203.285874, fps:440.07 imgs/sec, lr:0.0023212337400764227\n2021-08-05 16:09:13,947:INFO:epoch[1], iter[1800], loss:203.014775, fps:451.11 imgs/sec, lr:0.0024576964788138866\n2021-08-05 16:09:42,954:INFO:epoch[2], iter[1900], loss:194.683969, fps:441.28 imgs/sec, lr:0.0025941594503819942\n...\n```\n\n## [Evaluation Process](#contents)\n\n### Evaluation\n\nBefore running the command below, please check the checkpoint path used for evaluation. The file **yolov5.ckpt** used in the  follow script is the last saved checkpoint file, but we renamed it to \"yolov5.ckpt\".\n\n```python\n# run evaluation by python command\npython eval.py \\\n    --data_dir=xxx/dataset \\\n    --pretrained=xxx/yolov5.ckpt \\\n    --eval_shape=640 \u003e log.txt 2\u003e\u00261 \u0026\nOR\n# run evaluation by shell script\nbash run_eval.sh xxx/dataset xxx/yolov5.ckpt\n```\n\nThe above python command will run in the background. You can view the results through the file \"log.txt\". The mAP of the test dataset will be as follows:\n\n```python\n# log.txt\n=============coco eval reulst=========\nAverage Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.369\nAverage Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.573\nAverage Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.395\nAverage Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.218\nAverage Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.418\nAverage Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.482\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.298\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.501\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.557\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.395\nAverage Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.619\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.677\n2020-12-21 17:16:40,322:INFO:testing cost time 0.35h\n```\n\n## Inference Process\n\n### [Export MindIR](#contents)\n\n```shell\npython export.py --ckpt_file [CKPT_PATH] --file_name [FILE_NAME] --file_format [FILE_FORMAT]\n```\n\nThe ckpt_file parameter is required,\n`file_format` should be in [\"AIR\", \"MINDIR\"]\n\n### Infer on Ascend310\n\nBefore performing inference, the mindir file must be exported by `export.py` script. We only provide an example of inference using MINDIR model.\nCurrent batch_Size can only be set to 1.\n\n```shell\n# Ascend310 inference\nbash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [ANN_FILE] [DVPP] [DEVICE_ID]\n```\n\n- `DVPP` is mandatory, and must choose from [\"DVPP\", \"CPU\"], it's case-insensitive. The DVPP hardware restricts width 16-alignment and height even-alignment. Therefore, the network needs to use the CPU operator to process images.\n- `DATA_PATH` is mandatory, path of the dataset containing images.\n- `ANN_FILE` is mandatory, path to annotation file.\n- `DEVICE_ID` is optional, default value is 0.\n\n### result\n\nInference result is saved in current path, you can find result like this in acc.log file.\n\n```bash\nAverage Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.369\nAverage Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.573\nAverage Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.395\nAverage Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.218\nAverage Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.418\nAverage Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.482\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.298\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.501\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.557\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.395\nAverage Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.619\nAverage Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.677\n```\n\n# [Model Description](#contents)\n\n## [Performance](#contents)\n\n### Evaluation Performance\n\nYOLOv5 on 118K images(The annotation and data format must be the same as coco2017)\n\n| Parameters                 | YOLOv5s                                                      | YOLOv5s                                                      |\n| -------------------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |\n| Resource                   | Ascend 910 ；CPU 2.60GHz，192cores; Memory, 755G               | GPU NV SMX2 V100-32G                                         |\n| uploaded Date              | 7/12/2021 (month/day/year)                                   | 9/15/2021 (month/day/year)                                   |\n| MindSpore Version          | 1.2.0                                                        | 1.3.0                                                        |\n| Dataset                    | 118K images                                                  | 118K images                                                  |\n| 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      |\n| Optimizer                  | Momentum                                                     | Momentum                                                     |\n| Loss Function              | Sigmoid Cross Entropy with logits, Giou Loss                 | Sigmoid Cross Entropy with logits, Giou Loss                 |\n| outputs                    | boxes and label                                              | boxes and label                                              |\n| Loss                       | 111.970097                                                   | 85                                                           |\n| Speed                      | 8p about 450 FPS                                             | 8p about 290 FPS                                             |\n| Total time                 | 8p 21h28min                                                  | 8p 35h                                                       |\n| Checkpoint for Fine tuning | 53.62M (.ckpt file)                                          | 58.87M (.ckpt file)                                          |\n| Scripts                    | https://gitee.com/mindspore/models/tree/master/official/cv/yolov5 | https://gitee.com/mindspore/models/tree/master/official/cv/yolov5 |\n\n### Inference Performance\n\n| Parameters          | YOLOv5s                                        | YOLOv5s                                      |\n| ------------------- | -----------------------------------------------| ---------------------------------------------|\n| Resource            | Ascend 910 ；CPU 2.60GHz，192cores; Memory, 755G | GPU NV SMX2 V100-32G                         |\n| Uploaded Date       | 7/12/2021 (month/day/year)                     | 9/15/2021 (month/day/year)                   |\n| MindSpore Version   | 1.2.0                                          | 1.3.0                                        |\n| Dataset             | 20K images                                     | 20K images                                   |\n| batch_size          | 1                                              | 1                                            |\n| outputs             | box position and sorces, and probability       | box position and sorces, and probability     |\n| Accuracy            | mAP \u003e= 36.7%(shape=640)                        | mAP \u003e= 36.7%(shape=640)                      |\n| Model for inference | 56.67M (.ckpt file)                            | 58.87M (.ckpt file)                          |\n\n### Transfer Learning\n\n# [Description of Random Situation](#contents)\n\nIn dataset.py, we set the seed inside “create_dataset\" function. We also use random seed in train.py.\n\n# [ModelZoo Homepage](#contents)\n\n Please check the official [homepage](https://gitee.com/mindspore/models).  \n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fnemonameless%2Fmindsporeyolov5","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fnemonameless%2Fmindsporeyolov5","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fnemonameless%2Fmindsporeyolov5/lists"}