This file documents a large collection of baselines trained with pysot. All configurations for these baselines are located in the experiments directory. The tables below provide results about inference. Links to the trained models as well as their output are provided.
| Model （arch+backbone+xcorr） |
VOT2016 (EAO/A/R) |
VOT2018 (EAO/A/R) |
OTB2015 (AUC/Prec.) |
VOT2018-LT (F1) |
Speed (fps) |
Download |
|---|---|---|---|---|---|---|
| siamrpn_alex_dwxcorr | 0.393/0.618/0.238 | 0.352/0.576/0.290 | - | - | 180 | link |
| siamrpn_alex_dwxcorr_otb | - | - | 0.666/0.876 | - | 180 | link |
| siamrpn_r50_l234_dwxcorr | 0.464/0.642/0.196 | 0.415/0.601/0.234 | - | - | 35 | link |
| siamrpn_r50_l234_dwxcorr_otb | - | - | 0.696/0.914 | - | 35 | link |
| siamrpn_mobilev2_l234_dwxcorr | 0.455/0.624/0.214 | 0.410/0.586/0.229 | - | - | 75 | link |
| siammask_r50_l3 | 0.455/0.634/0.219 | 0.423/0.615/0.248 | - | - | 56 | link |
| siamrpn_r50_l234_dwxcorr_lt | - | - | - | 0.629 | 20 | link |
Note:
- speed tested on GTX-1080Ti
alexdenotes AlexNet,r50_lxyzdenotes the outputs of stage x, y, and z in ResNet-50, andmobilev2denotes MobileNetV2.dwxcorrdenotes Depth-wise Cross Correlation. See more in SiamRPN++ Section 3.4.- The suffixes
otbandltare designed for the OTB and VOT long-term tracking challenge, the default (without suffix) is designed for VOT short-term tracking challenge.
All models available for download through this document are licensed under the Creative Commons Attribution-ShareAlike 3.0 license.