search

huanglianghua / siamfc-pytorch

A clean PyTorch implementation of SiamFC tracking/training, evaluated on 7 datasets.
MIT License
628 stars 134 forks source link

Questions about test accuracy #15

Open SaintLogos1234 opened 5 years ago

SaintLogos1234 commented 5 years ago

I used your code to train with VID data set, and then tested the trained model. I found that the test results on OTB50 and VOT2018 were quite different from the results you gave, my test results: OTB:SR:0.517, pr:0.698 VOT2018: "speed_fps": 74.3267303229599, "robustness": 60.76630229928382, "accuracy": 0.4865348029482427

dkgupta90 commented 5 years ago

I too tested it on UAV123 and get very different precision and success scores. Any pointers on this?

sansanfree commented 5 years ago

I too tested it on UAV123 and get very different precision and success scores. Any pointers on this? Did you train the model on got-10k datasets? i found that the results with my own training model on TC128 and OTB2015 are worse than the results provided by authors?

dkgupta90 commented 5 years ago

Hi @SaintLogos1234 and @sansanfree I was wondering if you figure out the issue. I am still struggling and would have to stop using this toolkit if the results don't match. Looking forward to hear from you guys.

dkgupta90 commented 5 years ago

I too tested it on UAV123 and get very different precision and success scores. Any pointers on this? Did you train the model on got-10k datasets? i found that the results with my own training model on TC128 and OTB2015 are worse than the results provided by authors? I had trained on ISLRVC-VID dataset as has also been used by the author.

huanglianghua commented 5 years ago

The code have been updated with some modifications that make the performance more stable (when training on GOT-10k, AUC on OTB-15 is stable at around 0.57~0.60).

xuwei1245645564 commented 5 years ago

How can I do the AUC test? There is no evaluation code in this repository.

HonglinChu commented 4 years ago

I too tested it on UAV123 and get very different precision and success scores. But the results on OTB2015 is similar to yours. i dont know why