yixchen / yourefit_eru Goto Github PK

Thank you for your excellent work!
I tried to reproduce the performance in the paper, but the results were poor. I trained the model directly using a single GPU by

python train.py --data_root ./ln_data/ --dataset yourefit --gpu gpu_id --use_paf --use_sal --large

and the evaluation results are:

Accuracy = 0.18784972022382093
Small Accuracy = 0.0 in 418.0 samples
Medium Accuracy = 0.005063291139240506 in 395.0 samples
Large Accuracy = 0.5319634703196348 in 438.0 samples
Accuracy = 0.07993605115907274
Small Accuracy = 0.0 in 418.0 samples
Medium Accuracy = 0.0 in 395.0 samples
Large Accuracy = 0.228310502283105 in 438.0 samples
Accuracy = 0.019184652278177457
Small Accuracy = 0.0 in 418.0 samples
Medium Accuracy = 0.0 in 395.0 samples
Large Accuracy = 0.0547945205479452 in 438.0 samples

I'd like to know if there are any other settings needed?

I've encountered an issue with the YouRefIt Request form - it is no longer accepting responses.
If this is an error, could you please look into it?

Did you normalize the bounding box to train the model (i.e. calculate the loss for training)? If you normalize the bounding box, did you unnormalize the bounding box to calculate IOU? Thanks in advance.

When running python evaluate_results.py, I only get the

I want to get the bounding box, can you help me? Thanks!