wenguanwang / ags Goto Github PK

Perfect work!!! I notice that you have added the results on DAVIS2017. If it is possible, could you talk about how you modify this model to adapt instance-level object segmentation?

你好，你的模型权重下载不了，想问一下模型权重多大啊

What line of code do you use for convLSTM?

请问这个算法适用有多个运动目标的情况吗？

How much time dose this model process a frame?
What's your fps?

非常感谢您的工作，发现您在百度网盘的model和数据已经不见了，可以在再上传一下吗，谢谢。

Our model, segmentation and saliency results, and fixation data can also be found at Baidu Pan https://pan.baidu.com/s/1hz6iVOAMW3QRORebK3mC3Q , extaction code: 6prn

I try to experiment on the FBMS59 dataset.
I use your results on FBMS and davis16 matlab code to compute evaluation metric J and
obtain 73.95 (76.0 in your paper).

The groundtruth for fbms is generated from downloaded official dataset.

        if len(label.shape) ==3:
            label_1 = label[:,:,0]
            label_1 = (label_1==0).astype(np.uint8)

            label_2 = label[:,:,1]
            label_2 = (label_2==0).astype(np.uint8)

            label_3 = label[:,:,2]
            label_3 = (label_3==0).astype(np.uint8)
            label = label_1 + label_2 + label_3
            # label = label_1
            label = (label>0).astype(np.uint8)
            # label = label
        else:
            label = (label==255).astype(np.uint8)

it is the code for binary mask generation. I want to ask what is the question for my gt generation?
Can you give some help for this? It will be so thankful!!!!