liudaizong / f2net

Code for AAAI 2021 paper: F2Net: Learning to Focus on the Foreground for Unsupervised Video Object Segmentation

Hi, I'm Daizong Liu 👋

• 🔭 I’m currently working toward the Ph.D. degree at Wangxuan Institute of Computer Technology of Peking University.
• 🌱 My research interests include multi-modal learning, 3D adversarial attacks, etc.

f2net's Issues

trained model

Hi, this work is very interesting. I want to try your model to some videos, but I can not find any finished-trained model on the github. Could you upload it? Thanks in advance.

About trianing

Thanks for your great work!!!
I have a question about the training strategy. The paper says "During training the Center Guiding Appearance Diffusion
module, we first separately train the Center Prediction Branch and Spatial-Prior Guided Appearance Matching in parallel in the first 20 epochs, where we feed the appearancematching module with the center point of ground truth." But I could not find the code for training the Center Prediction Branch and Spatial-Prior Guided Appearance Matching. Does ie means that for the first 20 epochs, the code below is used to train the network? Or the code is somewhere else but I didn't find. Could you please locate it ?
`if i_iter%3 ==0: #对于静态图片的训练

            pred1, temp, heat = model(images, images, images, img_ct, img_guass)
            heat = _sigmoid(heat)
            loss = 0.1*(loss_calc1(temp, labels) + 0.8* loss_calc2(temp, labels))+ loss_calc3(heat, img_guass)
            loss.backward()

`

About FBMS evaluation

Thank you very much for sharing your excellent work, and I would like to ask two questions about the FBMS dataset. The first is how do you deal with multiple object sequences in the test set, and do you treat them all as foreground segmentation? Second, did you use the Official DAVIS evaluation code to evaluate the FBMS dataset?

about test

Thank you for your job on video segmentation. Now I try to follow your work, and can i get your test.py for testing the trained model. Thanks again.

License?

Could you specify the license of your code? Preferrably MIT