search

researchmm / Stark

[ICCV'21] Learning Spatio-Temporal Transformer for Visual Tracking
MIT License
649 stars 143 forks source link

Hello!About the track speed evaluation #67

Open davidyang180 opened 2 years ago

davidyang180 commented 2 years ago

Hello!The running speed of the model in the paper can reach 30~40fps,but when I download the pre-trained model and test it on got10k dataset, tracker.py set time stamps before and after tracking a single frame by referring to the time module, but it is not as fast as described in the paper. Isn't the evaluation of tracking time calculated in this way?

MasterBin-IIAU commented 2 years ago

@davidyang180 Hi, to speed up the evaluation on the tracking benchmarks, we run 32 processes in parallel. This can save the "total evaluation time" but would reduce the "speed per sequence". When testing speed, we actually only takes the forward propagation time in account.