Mismatch in the result (AP)

[jayaramreddy10]

Hi,

I have a query on test_seen result. When I evaluate kinect checkpoint on test_seen dataset, I get much lesser AP compared to what's reported. Could you please let me know if I am executing it wrong?

Command: CUDA_VISIBLE_DEVICES=0 python test.py --camera kinect --dump_dir logs/logs_kn --checkpoint_path ckpts/minkuresunet_kinect.tar --batch_size 1 --dataset_root data --infer --eval --collision_thresh -1

Result:

Also, I was using the pretrained weights from this github (https://github.com/graspnet/graspness_unofficial). If these are not the correct ones, could you please share the correct files to mailid: ramreddyai1010@gmail.com.

[wanghq-thu]

Hi,

I have a query on test_seen result. When I evaluate kinect checkpoint on test_seen dataset, I get much lesser AP compared to what's reported. Could you please let me know if I am executing it wrong?

Command: CUDA_VISIBLE_DEVICES=0 python test.py --camera kinect --dump_dir logs/logs_kn --checkpoint_path ckpts/minkuresunet_kinect.tar --batch_size 1 --dataset_root data --infer --eval --collision_thresh -1

Result:

Also, I was using the pretrained weights from this github (https://github.com/graspnet/graspness_unofficial). If these are not the correct ones, could you please share the correct files to mailid: ramreddyai1010@gmail.com.

Hello, I have also encountered a similar situation. The accuracy is only a little over 10 points. Have you made any progress now. Have you trained the model yourself.

[jayaramreddy10]

Hey, I tried with the checkpoint from this repo. (https://github.com/user432/gamma). And it gave me AP of close to 0.55 to 0.60.