leonzfa
leonzfa
@vinitra Yes. I used the onnx-coreml for my conversion. But in fact there is no Identity layer in the example code (Keras to ONNX Conversion). Why this onnxmltools generates Identity...
@zhangguanghui1 Hi, I have updated the code. Please following the instructions in README.md.I will keep on checking the code. If there is still any problem, I will fix the problems...
@zhangguanghui1 It seems to be the problem from lmdb. Have you downloaded the Scene Flow dataset? Anyway, if you have downloaded all the datasets, please use Terminal to run "sh...
@zhangguanghui1 You don't need to download all the datasets. But you should modify the scripts (train_rob_stage_one.prototxt and train_rob_stage_two.prototxt), and remove the "CustomData" layers that used to read data from other...
@TrackingBird I am sorry but I haven't test the code on Windows. Since there is Windows version of Caffe, so I think if you can correctly compile the Correlation1d and...
@catta202000 Their are two networks here. One is used in the CVPR paper, which corresponds to "deploy_iresnet.tpl.prototxt". The other is used in the ROB 2018, which corresponds to "deploy_iResNet_ROB.tpl.prototxt". The...
You can set the MAX_SIZE smaller, e.g., MAX_SIZE = 800000. Or you can uncomment line 180 to line 190 in file test_rob.py, which is used to calculate the runtime of...
@kyuhyoung When you submit your results to Middlebury benchmark, you should provide both left disparity and right disparity. This is only required by Middlebury benchmark. For KITTI and ETH3D, only...
@zmzhaoming 这个问题我没遇到过，你可以试试单步调试进去，看看哪里有不正确的，是不是图片读错了。。。或者你注释掉get_image_size那两行，然后手动设置一下图片的大小。。