I observed a significant difference in the model's output when using different batch sizes. Specifically:
- Left: Output with
batchsize=16. - Right: Output with
batchsize=6.
I don't know why this discrepancy occurs. I tested multiple datasets, and this phenomenon occurs in all of them. Any insights or suggestions are appreciated.
python train.py --train_image_path {train_image_path} --train_mask_path {train_mask_path} --save_path {save_path} --epoch 50 --gpu_choose 0 --batch_size 16 --num_workers 4
python test.py --checkpoint {checkpoint_path} --test_image_path {img_path} --test_gt_path {label_path} --save_path {save_path} --device 0
python eval.py --dataset_name {dataset_name} --pred_path {pred_path} --gt_path {gt_path}