Cross-Camera-Multi-Target-Vehicle-Tracking-Competition

AI-Driven Future of Transportation: Cross-Camera Multi-Target Vehicle Tracking Competition – Model Development Session

Progress

• 28/05/2024 - Inference on test data (AI_CUP_testdata/)
• 27/04/2024 - Change loss function for ReID module (Triplet to Circle)
• 22/04/2024 - Evaluate on YOLOv8 and YOLOv9-E and train ReID with Imgsz=960 (weird results)
• 17/04/2024 - Evaluate on YOLOv7-E6E and train YOLOv9-E (17/04/2024 - 22/04/2024)
• 14/04/2024 - Setup and Train YOLOv7-E6E with ReID (14/04/2024 - 16/04/2024)

Hardware Information

• CPU: AMD Ryzen 5 5600X 6-Core @ 12x 3.7GHz
• GPU: NVIDIA GeForce RTX 3060 Ti (8G)
• RAM: 48087MiB

Create Conda Environment

$ conda create -n botsort python=3.7 -y
$ conda activate botsort

# https://pytorch.org/get-started/locally/

$ git clone https://github.com/ricky-696/AICUP_Baseline_BoT-SORT.git
$ cd AICUP_Baseline_BoT-SORT/
$ pip install numpy
$ pip install -r requirements.txt

# Install pycocotools
$ pip3 install cython; pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

# Cython-bbox
$ pip install cython_bbox

# faiss cpu / gpu
$ pip install faiss-cpu
$ pip install faiss-gpu

Folder Structure

MCMOT/
    ├── AICUP_Baseline_BoT-SORT/
    └── datasets/
        └── train/

Prepare ReID Dataset

$ cd AICUP_Baseline_BoT-SORT/

$ python fast_reid/datasets/generate_AICUP_patches.py --data_path ../datasets/train
# output: /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/fast_reid/datasets/AICUP-ReID/

Prepare YOLOv7 Dataset

$ cd AICUP_Baseline_BoT-SORT/

$ python yolov7/tools/AICUP_to_YOLOv7.py --AICUP_dir ../datasets/train --YOLOv7_dir datasets/AI_CUP_MCMOT_dataset/yolo
# output: /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/datasets/AI_CUP_MCMOT_dataset/yolo

Download Pretrained Weight

$ cd AICUP_Baseline_BoT-SORT/
$ mkdir pretrained
$ cd pretrained/
$ wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7_training.pt
$ wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-e6e_training.pt

Train the ReID Module for AICUP

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 25: IMS_PER_BATCH: 60    # 256

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

The training results are stored by default in logs/AICUP/bagtricks_R50-ibn.

The storage location and model hyperparameters can be modified in fast_reid/configs/AICUP/bagtricks_R50-ibn.yml.

You can refer to fast_reid/fastreid/config/defaults.py to find out which hyperparameters can be modified.

Fine-tune YOLOv7 for AICUP

• The dataset path is configured in yolov7/data/AICUP.yaml.

  # train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
  train: /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/datasets/AI_CUP_MCMOT_dataset/yolo/train
  val: /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/datasets/AI_CUP_MCMOT_dataset/yolo/val
  
  # number of classes
  nc: 1
  
  # class names
  names: [ 'car' ]
  ```https://github.com/wish44165/Cross-Camera-Multi-Target-Vehicle-Tracking-Competition/blob/main/assets/v9-e2_circleLoss.png
  

• The model architecture can be configured in yolov7/cfg/training/yolov7-AICUP.yaml.
• Training hyperparameters are configured in yolov7/data/hyp.scratch.custom.yaml (default is yolov7/data/hyp.scratch.p5.yaml).

$ cd AICUP_Baseline_BoT-SORT/

# official
## finetune p5 models
$ python yolov7/train.py --device 0 --batch-size 16 --epochs 50 --data yolov7/data/AICUP.yaml --img 1280 1280 --cfg yolov7/cfg/training/yolov7-AICUP.yaml --weights 'pretrained/yolov7-e6e.pt' --name yolov7-AICUP --hyp data/hyp.scratch.custom.yaml
## finetune p6 models
$ python yolov7/train_aux.py --device 0 --batch-size 16 --epochs 50 --data yolov7/data/AICUP.yaml --img 1280 1280 --cfg yolov7/cfg/training/yolov7-w6-AICUP.yaml --weights 'pretrained/yolov7-e6e.pt' --name yolov7-w6-AICUP --hyp data/hyp.scratch.custom.yaml

$ python yolov7/train.py --device 0 --batch-size 1 --epochs 50 --data yolov7/data/AICUP.yaml --img 1280 1280 --cfg yolov7/cfg/training/yolov7-AICUP.yaml --weights 'pretrained/yolov7-e6e_training.pt' --name yolov7-AICUP --hyp data/hyp.scratch.custom.yaml

Tracking and creating the submission file for AICUP (Demo)

$ cd AICUP_Baseline_BoT-SORT/

# Track one <timestamp> with BoT-SORT(-ReID) based YOLOv7 and multi-class (We only output class: 'car').
$ python3 tools/mc_demo_yolov7.py --weights runs/train/yolov7-AICUP/weights/best.pt --source /home/wish/pro/AICUP/MCMOT/datasets/train/images/0902_150000_151900 --device "0" --name "0902_150000_151900" --fuse-score --agnostic-nms --with-reid --fast-reid-config fast_reid/configs/AICUP/bagtricks_R50-ibn.yml --fast-reid-weights logs/AICUP_115/bagtricks_R50-ibn/model_0058.pth

$ Track all <timestamps> in the directory, you can execute the bash file we provided.
$ bash tools/track_all_timestamps.sh --weights runs/train/yolov7-AICUP/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config "fast_reid/configs/AICUP/bagtricks_R50-ibn.yml" --fast-reid-weights logs/AICUP_115/bagtricks_R50-ibn/model_0058.pth

Evaluate your submission

$ cd AICUP_Baseline_BoT-SORT/

# Before evaluation, you need to run tools/datasets/AICUP_to_MOT15.py to convert ground truth into submission format:
$ python tools/datasets/AICUP_to_MOT15.py --AICUP_dir /home/wish/pro/AICUP/MCMOT/datasets --MOT15_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15

# You can use tools/evaluate.py to evaluate your submission by the following command:
$ cp 09*/*.txt tracking_results/
$ cp 10*/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/tracking_results

YOLOv8

$ cd AICUP_Baseline_BoT-SORT/

$ conda activate yuhs1
$ pip install ultralytics

$ git clone https://github.com/ultralytics/ultralytics.git
$ cd ultralytics/

# train
$ wget https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8x.pt
$ python train.py

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v8.sh --weights ./ultralytics/runs/detect/train/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v8x_128/tracking_results/

YOLOv8 with default ReID training

YOLOv8x with ReID imgsz=128 and CircleLoss training (2nd trial)

YOLOv9

$ cd AICUP_Baseline_BoT-SORT/

$ git clone https://github.com/WongKinYiu/yolov9.git
$ cd yolov9/
$ pip install seaborn thop
$ pip install ipython
$ pip install psutil

# demo
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-c-converted.pt
$ python detect.py --source './data/images/horses.jpg' --img 640 --device 0 --weights './yolov9-c-converted.pt' --name yolov9_c_640_detect

# train
$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-e-converted.pt
$ python train_dual.py --workers 8 --device 0 --batch 1 --data data/AICUP.yaml --img 1280 --cfg models/detect/yolov9-e.yaml --weights './yolov9-e-converted' --name yolov9-e --hyp hyp.scratch-high.yaml --min-items 0 --epochs 50 --close-mosaic 4

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config "fast_reid/configs/AICUP/bagtricks_R50-ibn.yml" --fast-reid-weights logs/AICUP_115/bagtricks_R50-ibn/model_0058.pth

# Evaluate your submission
$ cp 09*/*.txt tracking_results/
$ cp 10*/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/tracking_results

YOLOv9 with circle loss

$ cd AICUP_Baseline_BoT-SORT/

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config "fast_reid/configs/AICUP/bagtricks_R50-ibn.yml" --fast-reid-weights logs/AICUP_115/bagtricks_R50-ibn/model_0048.pth

# Evaluate your submission
$ cp 09*/*.txt tracking_results/
$ cp 10*/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_circleloss/tracking_results/

---

Train the ReID Module for AICUP (imgsz=960: weird results)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [960, 960]    # [256, 256]
>> line 5: SIZE_TEST: [960, 960]    # [256, 256]
>> line 25: IMS_PER_BATCH: 4    # 256
>> line 34: IMS_PER_BATCH: 960    # 256

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

The training results are stored by default in logs/AICUP/bagtricks_R50-ibn.

The storage location and model hyperparameters can be modified in fast_reid/configs/AICUP/bagtricks_R50-ibn.yml.

You can refer to fast_reid/fastreid/config/defaults.py to find out which hyperparameters can be modified.

Train the ReID Module for AICUP (imgsz=704, w/ circleLoss: weird results)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [704, 704]    # [256, 256]
>> line 5: SIZE_TEST: [704, 704]    # [256, 256]
>> line 25: IMS_PER_BATCH: 8    # 256
>> line 34: IMS_PER_BATCH: 704    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_704_circleLoss/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_704_circleLoss/model_0058.pth

# Evaluate your submission
$ cp 09*/*.txt tracking_results/
$ cp 10*/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_704_circleLoss/tracking_results/

Train the ReID Module for AICUP (imgsz=320, w/ IBN, NL, BNneck, EMA, CircleLoss)

• output folder: bagtricks_R50-ibn_320/

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [320, 320]    # [256, 256]
>> line 5: SIZE_TEST: [320, 320]    # [256, 256]
>> line 25: IMS_PER_BATCH: 24    # 256
>> line 34: IMS_PER_BATCH: 320    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_320/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_320/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_320/tracking_results/

Train the ReID Module for AICUP (imgsz=256, w/ IBN, NL, BNneck, EMA, CircleLoss)

• output folder: bagtricks_R50-ibn_256_v2/

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [256, 256]    # [256, 256]
>> line 5: SIZE_TEST: [256, 256]    # [256, 256]
>> line 25: IMS_PER_BATCH: 60    # 256
>> line 34: IMS_PER_BATCH: 256    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_256_v2/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_256_v2/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_256_v2/tracking_results/

Train the ReID Module for AICUP (imgsz=224, w/ circleLoss)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [224, 224]    # [256, 256]
>> line 5: SIZE_TEST: [224, 224]    # [256, 256]
>> line 25: IMS_PER_BATCH: 68    # 256
>> line 34: IMS_PER_BATCH: 224    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_224/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_224/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_224/tracking_results/

Train the ReID Module for AICUP (imgsz=192, w/ circleLoss)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [192, 192]    # [256, 256]
>> line 5: SIZE_TEST: [192, 192]    # [256, 256]
>> line 25: IMS_PER_BATCH: 84    # 256
>> line 34: IMS_PER_BATCH: 192    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_192/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_192/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_192/tracking_results/

Train the ReID Module for AICUP (imgsz=160, w/ circleLoss)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [160, 160]    # [256, 256]
>> line 5: SIZE_TEST: [160, 160]    # [256, 256]
>> line 25: IMS_PER_BATCH: 128    # 256
>> line 34: IMS_PER_BATCH: 160    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_160/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_160/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_160/tracking_results/

Train the ReID Module for AICUP (imgsz=128, w/ circleLoss)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [128, 128]    # [256, 256]
>> line 5: SIZE_TEST: [128, 128]    # [256, 256]
>> line 25: IMS_PER_BATCH: 256    # 256
>> line 34: IMS_PER_BATCH: 128    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_128/tracking_results/

Train the ReID Module for AICUP (imgsz=96, w/ circleLoss)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [96, 96]    # [256, 256]
>> line 5: SIZE_TEST: [96, 96]    # [256, 256]
>> line 25: IMS_PER_BATCH: 480    # 256
>> line 34: IMS_PER_BATCH: 96    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_96/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_96/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_96/tracking_results/

Train the ReID Module for AICUP (imgsz=64, w/ circleLoss)

fast_reid/configs/AICUP/bagtricks_R50-ibn.yml

>> line 4: SIZE_TRAIN: [64, 64]    # [256, 256]
>> line 5: SIZE_TEST: [64, 64]    # [256, 256]
>> line 25: IMS_PER_BATCH: 800    # 256
>> line 34: IMS_PER_BATCH: 64    # 256

fast_reid/configs/Base-bagtricks.yml

>> line 22: NAME: ("CrossEntropyLoss", "CircleLoss",)    # ("CrossEntropyLoss", "TripletLoss",)

$ cd AICUP_Baseline_BoT-SORT/

$ python3 fast_reid/tools/train_net.py --config-file fast_reid/configs/AICUP/bagtricks_R50-ibn.yml MODEL.DEVICE "cuda:0"

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_64/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_64/model_0058.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e_64/tracking_results/

---

Data augmentation

• original: train:val = 23307:8640 (0.3707040803192174)

$ cd AICUP_Baseline_BoT-SORT/
$ git clone https://github.com/Paperspace/DataAugmentationForObjectDetection.git
$ python data_aug.py

• train:val = 39240:14032 (0.35759429153924566)

YOLOv9 (e2)

$ cd AICUP_Baseline_BoT-SORT/yolov9/

$ wget https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-e.pt

# train
$ python train_dual.py --workers 8 --device 0 --batch 1 --data data/AICUP.yaml --img 1280 --cfg models/detect/yolov9-e.yaml --weights ./yolov9-e.pt --name yolov9-e --hyp hyp.scratch-high.yaml --min-items 0 --epochs 10 --close-mosaic 2

# Tracking and creating the submission file for AICUP
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e2/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/datasets/train/images --device "0" --fast-reid-config "fast_reid/configs/AICUP/bagtricks_R50-ibn.yml" --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_circleLoss/model_0048.pth

# Evaluate your submission
$ cp *00/*.txt tracking_results/
$ python tools/evaluate.py --gt_dir /home/wish/pro/AICUP/MCMOT/datasets_MOT15 --ts_dir /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/runs/detect/v9-e2_circleLoss/tracking_results/

---

Inference

• inference.ipynb

# yolov8 128 (0.828468)
$ bash tools/track_all_timestamps_v8.sh --weights ./ultralytics/runs/detect/train/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/32_33_AI_CUP_testdataset/AI_CUP_testdata/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/model_0058.pth
$ cd runs/detect/
$ cp 09*/*.txt 2nd/
$ cp 10*/*.txt 2nd/

# yolov9_e2 256_v2 (0.824332)
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e2/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/32_33_AI_CUP_testdataset/AI_CUP_testdata/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_256_v2/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_256_v2/model_0058.pth
$ cd runs/detect/
$ cp 09*/*.txt 4th/
$ cp 10*/*.txt 4th/

# yolov9_e2 128 (0.823188)
$ bash tools/track_all_timestamps_v9.sh --weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/yolov9/runs/train/yolov9-e2/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/32_33_AI_CUP_testdataset/AI_CUP_testdata/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_128/model_0058.pth
$ cd runs/detect/
$ cp 09*/*.txt 5th/
$ cp 10*/*.txt 5th/

# yolov8 256_v2 (public: 0.828977 - 23rd	/ private: 0.945747088 - 18th)
$ bash tools/track_all_timestamps_v8.sh --weights ./ultralytics/runs/detect/train/weights/best.pt --source-dir /home/wish/pro/AICUP/MCMOT/32_33_AI_CUP_testdataset/AI_CUP_testdata/images --device "0" --fast-reid-config /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_256_v2/config.yaml --fast-reid-weights /home/wish/pro/AICUP/MCMOT/AICUP_Baseline_BoT-SORT/logs/AICUP_115/bagtricks_R50-ibn_256_v2/model_0058.pth
$ cd runs/detect/
$ cp 09*/*.txt 6th/
$ cp 10*/*.txt 6th/

---

Acknowledgements

• AICUP Baseline: BoT-SORT (Prev)
• Official BoT-SORT
• Official YOLOv7
• Official YOLOv8
• Official YOLOv9
• Data Augmentation For Object Detection

References

v7

• What is different in P5 model and P6 model? #141

v9

• train.py tran_dual.py train_triple.py The relationship and difference between the three #1
• What's the difference between yolov9-c-converted.pt and gelan-c.pt? #131
• Converted version, e and c and gelan? #209
• Hello, what is the difference between yolov9-c-converted and yolov9-c? #326