label.sh

#!/usr/bin/env bash
#
# Slurm arguments
#
#SBATCH --cpus-per-gpu=2        # Number of CPU cores to allocate
#SBATCH --export=ALL
#SBATCH --gres=gpu:4             # Number of GPU's
#SBATCH --mem-per-gpu=20G	 # Memory to allocate in MB per allocated CPU core
#SBATCH --partition=gpu

echo "Start labeling"

conda activate astod

PORT=$(( $RANDOM % 50 + 10000 ))

torchrun --nproc_per_node=4 --master_addr 127.0.0.5 --master_port $PORT src/label.py \
         --config configs/faster_rcnn_coco_teacher.yaml \
         --train-file coco/annotations/instances_train2017.$1@$2.json  \
         --unlabeled-file coco/annotations/instances_train2017.$1@$2-unlabeled.json \
         --output-dir coco/annotations/pseudo_labels/faster_rcnn/$2/teacher/model_$1/ \
         --checkpoint models/faster_rcnn/coco/$2/teacher/model_$1/best_model.pth \