hahv / yolov5_ai_class

object detection with yolov5 for mobility (vehicles) dataset in UAV AI class Spring Sem 2022

HOW TO THESE YOLOV5 CODE WITH MOBILITY DATASET

Find out origin work YOLOV5 here

• Step 1: Install requirement described in requirements.txt file

• Step 2: In the folder data, create a folder named mobility. Download and copy the train/valid/test data into that folder.

  • For mobility dataset (15 classes), data can be downloaded from this link
  • For mobility dataset (4 classes of group 2: Micro Car, SUV, Bus, Van), data can be downloaded from this link
  • For both cases, copy file mobility.yaml (in downloaded data) and put it into data folder
  • To train for 4 classes, edit the mobility.yaml

• Step 3: For training, run the code in the file zrun.bat with command prompt (terminal in macOS) - edit the batch, epoch, etc. (the argument --img 640 is fixed for pretrained model yolov5l). In case ./zrun.bat has Permission denied error, give the permission by running chmod +x zrun.bat and then ./zrun.bat. You can change the data augmentation configuration in file data/hyps/hyp.scratch-low.yaml

• Step 4: For run detection, run the code in the file zdetect.bat with command prompt (terminal in macOS) - edit the --conf threshold and the path for image folder. For run detection on video, run the code in the file zvideo.bat with command prompt (terminal in macOS) - edit the --conf threshold and the path for video/ video folder.

• Step 5: Check the results (for training or detect) inside folder runs

NOTE:You can download pretrained models (for 4/15 classes) from here