zosel260
zosel260
Thank you for kind reply.
I found out that there is no loading of net model. By enabling call of 'load_cnn' function from line 92 of init_features.m like below, I can run vgg model normally....
@yjkupup Please look over the tune_vot.py. In this file, you can see the codes like below: before running the tracking for one video, it saves 'Occ' string on the result...
@kongbia This codes must be remained because pos means positive samples and pos exists during training also.
@yjkupup I tried training the base and refinement module twice. But the final EAOs were only 0.350 and 0.356 on VOT2018. I think the only different things from author's were...
@yjkupup My cuda version is 10.1. I'm not sure nn.Upsample is problem but I don't think it has critical effect. Did you test other data set like VOT2019, I tested...
@yjkupup thanks!
@yjkupup Did you check the evaluation functions( utils/pysot/evaluation/ eao_benchmark.py) are proper tools for VOT metric?
after training, my final results was only EAO=0.35, my batch size were: - train base model : 64(e1 ~ 10), 16(e11 ~ 20) - train refinement : 32(e1 ~ 20)...
@SamihaSara From tools/test.py, you can see codes like: ``` if refine_enable: mask = net.track_refine((delta_y, delta_x)).to(device).sigmoid().squeeze().view( p.out_size, p.out_size).cpu().data.numpy() ``` in this codes, mask is the segmentation result obtained from the refinement...