search

researchmm / SiamDW

[CVPR'19 Oral] Deeper and Wider Siamese Networks for Real-Time Visual Tracking
http://openaccess.thecvf.com/content_CVPR_2019/html/Zhang_Deeper_and_Wider_Siamese_Networks_for_Real-Time_Visual_Tracking_CVPR_2019_paper.html
MIT License
751 stars 180 forks source link

Tiger1 from OTB #88

Open ISosnovik opened 4 years ago

ISosnovik commented 4 years ago

I found that in your OTB2013.json file you have ground truth for Tiger1 which is different from the official one. First of all, it contains 346 lines for 349 frames. While the original sequence has 354 frames. Where did you get it? Why and how did you generate OTB13/15 json files?

JudasDie commented 4 years ago

I found that in your OTB2013.json file you have ground truth for Tiger1 which is different from the official one. First of all, it contains 346 lines for 349 frames. While the original sequence has 354 frames. Where did you get it? Why and how did you generate OTB13/15 json files?

Hi, thanks for your interest first. The json file is generated according to the official Matlab toolkit. The 1-5 frames of tiger1 are skipped in the official toolkit because of occlusion or other factors. Please check "initOmit" in the official Matlab toolkit. I would suggest you debug the official Matlab toolkit if you want to know the detailed testing procedure in OTB. Besides, one easy way to verify the results on the official toolkit is to transfer the ".txt" files to the ".mat" format and evaluate it in the official toolkit.

ISosnovik commented 4 years ago

Could you share a link to the toolkit? I can't find it.

ISosnovik commented 4 years ago

@JudasDie