I have some problems with the FFb6d code, in fact I have been trying for a few days to reproduce the results/metrics indicated on the paper.
I first ran the code with the pre-trained network provided by you, and got the desired results of course.
I then launched a local training on my machine (without any change brought to the code, I have only GitClone the FFb6d directory) and I obtain the following results:
Did I do something wrong during the git clone? Or did I forget a part? Because the metrics as we can see are below what it could/should be. Do you have an idea of what I can do to correct this problem? Or where this problem could come from.
Thanks in advance for your helps and your response
Hello,
I have some problems with the FFb6d code, in fact I have been trying for a few days to reproduce the results/metrics indicated on the paper. I first ran the code with the pre-trained network provided by you, and got the desired results of course. I then launched a local training on my machine (without any change brought to the code, I have only GitClone the FFb6d directory) and I obtain the following results:
kps_pth in get_kps: datasets/linemod/kps_orb9_fps/ape_8_kps.txt loss_rgbd_seg 0.002694820675089778 loss_kp_of 0.4796774632306326 loss_ctr_of 0.037936181447335654 loss_all 0.5230032857826777 loss_target 0.5230032857826777 acc_rgbd 99.71056330771674 obj_id: 1 0.1 diameter: 0.010209865663 1 ape add auc: 63.393883012269 adds auc: 90.00338269221461 add(-s) auc: 63.393883012269 add < 0.1 diameter: 2.1904761904761907 ***adds < 0.1 diameter: 85.90476190476191
Use time: 422.245965719223 s
Did I do something wrong during the git clone? Or did I forget a part? Because the metrics as we can see are below what it could/should be. Do you have an idea of what I can do to correct this problem? Or where this problem could come from.
Thanks in advance for your helps and your response