Flow-Guided-Feature-Aggregation
Flow-Guided-Feature-Aggregation copied to clipboard
in resnet_v1_101_flownet_rfcn.py
line 910 using select_conv_feat = weight1 * warp_conv_feat_1 + weight2 * warp_conv_feat_2 seems just accumulate features from nearby frames. Paper said including its own, but in this step, it doesn't include its own feature map. Or is there anything I missed?
line 910 using select_conv_feat = weight1 * warp_conv_feat_1 + weight2 * warp_conv_feat_2 seems just accumulate features from nearby frames. Paper said including its own, but in this step, it doesn't include its own feature map. Or is there anything I missed?
Yeap,I notice these in training code(get_train_symbol func),have you figure out this?