Dear author,
I used the code you provided and ran this bash script on the IU Xray dataset for 70 epochs, but the performance metrics are far from those reported in the paper.
Here is my sh file, I modified the values for log_period, epoch, and batch_size, and used 7580 as the random seed. I didn't modify anything else. However, the results still don't reach reported in the paper.
python train_rl.py \
--image_dir ../../Code/data/iu_xray/images/ \
--ann_path ../../Code/data/iu_xray/annotation.json \
--dataset_name iu_xray \
--max_seq_length 60 \
--threshold 3 \
--batch_size 6 \
--epochs 70 \
--save_dir results/iu_xray \
--step_size 1 \
--gamma 0.8 \
--seed 7580 \
--topk 32 \
--beam_size 3 \
--log_period 50
Dear author,
I used the code you provided and ran this bash script on the IU Xray dataset for 70 epochs, but the performance metrics are far from those reported in the paper.
Here is my sh file, I modified the values for log_period, epoch, and batch_size, and used 7580 as the random seed. I didn't modify anything else. However, the results still don't reach reported in the paper.
python train_rl.py \ --image_dir ../../Code/data/iu_xray/images/ \ --ann_path ../../Code/data/iu_xray/annotation.json \ --dataset_name iu_xray \ --max_seq_length 60 \ --threshold 3 \ --batch_size 6 \ --epochs 70 \ --save_dir results/iu_xray \ --step_size 1 \ --gamma 0.8 \ --seed 7580 \ --topk 32 \ --beam_size 3 \ --log_period 50