search

ybsong00 / Vital_release

VITAL: VIsual Tracking via Adversarial Learning
BSD 2-Clause "Simplified" License
129 stars 37 forks source link

Is the offline pretraining code available? #6

Closed yiminglin-ai closed 5 years ago

yiminglin-ai commented 6 years ago

Dear Dr.Song, Thank you for open-sourcing the code. The VITAL paper is an excellent work and reading the paper is an inspiring experience. I noted that the pretrain folder in this repo does not include the code for the offline pretraining process, as you mentioned in the Model Initialization in Section 4 of your paper. In fact, the folder is identical to that in MDNet and does not involve adversarial training. I was wondering if you plan to open-source the training process, which I think would be super helpful for the community to understand better how to apply GAN to the tracking problem. Thanks for your time. Best regards, Yiming

Rheelt commented 6 years ago

Hi, @yl1991 it is my understanding that VITAL use pretrain MDNet, as for adversarial training, it takes place during the frist frame and online update phase.

yiminglin-ai commented 6 years ago

@coder-LeeTao Hi Lee, thanks for your answer. Below is some text I found in the paper

We initialize our model through a two-stage training. In the first step we offline pretrain the model using positive and negative samples from the training data, which is from [38]. In the second step we draw the samples from the first frame of the input sequence to finetune our model online. During offline pretraining, we randomly initialize D and perform the training in a few iterations, then we involve G for adversarial learning. See Sec. 3.1 for the details of the adversarial learning process where only positive samples are adopted.

I think the highlighted sentence states G is also involved in the offline training, what do you think?