ODCE

Project repository for the project object detection from clip embeddings for the EPFL course Visual Intelligence.

Dataset is saved at data/longclip-emb/text_features.pt & data/longclip-emb/vision_features.pt import detr is a bit tricky, I cloned the repo and added it to the path. git clone https://github.com/facebookresearch/detr.git

Be sure to rename the folder detr/datasets to something else, otherwise it conflicts with huggingface module "datasets"

Start an interactive session on a compute node (eg, 2 GPUs case):

bbox: Bounding box coordinates [x, y, w, h] of the annotated object.

srun -t 240 -A cs-503 --qos=cs-503 --gres=gpu:1 --mem=16G --pty bash

Then, on the compute node:

conda activate nanofms
wandb login
OMP_NUM_THREADS=1 torchrun --nproc_per_node=2 main.py