Imbalanced-Dataset-Project

Preparing

Prepare your data before training. The format of your data should follow the file in datasets. You can downlaod Cityscapes dataset and run
python get_bbox_seg.py
or download processed dataset here

Training stage

python train.py --dataroot ./dataset --name model_name --model seg2pix --which_model_netG unet_256  --lambda_A 100 --dataset_mode aligned --use_spp --no_lsgan --norm batch

Pretrained checkpoints put it under /checkpoints/model_name/

Testing stage

python test.py --dataroot data_path --name model_name --model seg2pix --which_model_netG unet_256   --dataset_mode aligned --use_spp --norm batch

Data augmentation stage

The results are under folder aug_pedestrian. Follow the instruction in MMdetection to convert Cityscapes to COCO fomat. Then run
python data_aug.py

Evaluation stage

Put the new Cityscapes dataset under mmdection folder .Follow the instruction in MMdetection to train a object detector and evaluate it

Acknowledgments

Heavily borrow the code from PSGAN