process_dataset.ipynb
It accepts the nyudv2.mat file from ./datasets folder
The output is .npy files corresponding to rgb, depth, raw-depth, instance and label map, in nyudv2 folder
hha.ipynb
Run the above python notebook in the directory ./hha_encoding folder
It saves the hha encoded file in the .npy format in nyudv2 folder
https://drive.google.com/file/d/1YFPXk6EvdhwYZZ9vLRLBbkmBpym6AzXA/view?usp=sharing
single images: https://drive.google.com/file/d/1y92N_EjhO96unAOfG5tBNkNSJFEeamIL/view?usp=sharing
all images combined: https://drive.google.com/file/d/1H6HzI1_xTMMxivb7zfG5t-BHcB_zk8J9/view?usp=sharing
- Exp1: finetune all layers of YOLOv5 for RGB images
- Exp2: freeze first 10 layers of YOLOv5 for RGB images
- Exp3: finetune all layers of YOLOv5 for HHA images
- Exp4: freeze first 10 layers of YOLOv5 for HHA images
- WandB results: https://wandb.ai/deepti-rawat/cv_project_yolov5?workspace=user-deepti-rawat