Pierre Baqué

Hello liangzhangxd, Thanks for your interest. You need to use our code to replace the original denseCRF used in [DeepLab ](https://bitbucket.org/deeplab/deeplab-public). All the details for running experiments are provided there....

Hello, I have the same problem currently, I don't manage to get reasonable results. Did you find out what was the issue? Best, Pierre

Hey Martin, Thanks for your answer. Therefore, in the post (https://github.com/torrvision/crfasrnn/issues/28), when you say that you get "better results", it is not compared to what the guys report in the...

Hi @bittnt, Could you maybe provide the pretrained version of FCN-8 that you used as a starting point? When can we expect the training scripts to be released? Thanks Pierre

OK, that is what I used. There must be an issue somewhere in the way I use the CRF part since that the initial results are far from being as...

Hello @martinkersner , I think I partially solved the issue. In you TVG_CRFRNN_COCO_VOC_TRAIN_20_CLASSES.prototxt and TVG_CRFRNN_COCO_VOC_TEST_20_CLASSES.prototxt, all the names have "-ft" at the end. It is not the case in https://gist.github.com/longjon/1bf3aa1e0b8e788d7e1d#file-readme-md,...

@bittnt Thank you, I'll have a look. Cheers

Hello, Thank you for your message. What are the parameters that you don't manage to set? What kind of results do you get? Does it look like the rectangle indices...

Hello, You need to calibrate your cameras in some way, using either a chessboard on the ground, or other markers. You need the intrinsic and extrinsic calibration values. Best, Pierre

Hello, Yes, it is normal. The inference part runs only on CPU but the other parts should be ran on GPU. Cheers