search

lizhou-cs / JointNLT

The official implementation for the CVPR 2023 paper Joint Visual Grounding and Tracking with Natural Language Specification.
MIT License
55 stars 2 forks source link

OTB_NL model for evalution #11

Open twotwo2 opened 1 year ago

twotwo2 commented 1 year ago

Thanks for your excellent work. I noticed that you provided the pre-trained models for evaluation: "JointNLT_OTB_NL.pth.tar" and "JointNLT_ep300.pth.tar". I think "JointNLT_OTB_NL.pth.tar" is specifically for evaluating on otb99 only via NL. I would like to know how the training strategy for this model differs from the "JointNLT_ep300.pth.tar" model. I would greatly appreciate it if I could get a reply from you.

lizhou-cs commented 1 year ago

Hello, thank you for your interest in our work. During our experiments, we found that the OTB dataset has unique labeling rules for bounding boxes that differ from other datasets. This resulted in poorer evaluation results with more training epochs for our model. Therefore, during the evaluation, we used the model trained for the 200th epoch using the same training process to assess the performance on the OTB dataset, while TNL2K and LaSOT utilized models trained for 300 epochs.

twotwo2 commented 11 months ago

Thank you for your reply. Another question I am puzzled is that otb has a relatively small amount of data, why is the training data sampled with a sampling ratio set otb99:lasot:tnl2k:ref = 1:1:1:1:1. May I ask if you have tried to change the sampling rate? I'm curious if increasing the sampling ratio of the lasot, tnl2k data improves the performance of the tracking?