AsCu

At the moment, I dont have much time to do it. You may try it yourself, which is not really difficult. You can just create a new class and does...

Hi, sorry for the late reply. `face_texture` actually describes the RGB of each vertex. The dimensions are the same as `vs` So if you use your own model, only make...

Could you provide more details? what is tensor.H in your case?

I think the timesteps are evenly distributed. Taking T = 1000 for example, if there are four timesteps to sample from for the student, it would be 999, 749, 499,...

Sorry for the late reply. The foreground seed cues are basically generated based on the foreground saliency. SALIENCY_TH_FG is the parameter that controls whether a pixel should be selected as...

Just edit the parameter in "config.py" in the "core" folder. As I said, I would recommend doing some ablation study here by simply removing those blank-array images. This would be...

Which backbone did you use, VGG or Resnet50? Did you use it for the VOC dataset? It is hard to tell from what you described. I do not know if...

I remember I was using 1.3.0 at that time. If you use your own dataset, I suggest you might dig up your dataset a little bit. For example, check the...

From what you said, it seems the issue comes more likely from your dataset than the toolbox itself. Cues are a very important starting point for the SEC framework, but...

It seems that the aspect ratio does have a huge impact on the results. At least in the original SEC paper, the ratio is not kept.