XuanyiLi

Yes, the current hitnet code performance of the pytorch version is 2.48 on [kitti.](http://www.cvlibs.net/datasets/kitti/eval_scene_flow_detail.php?benchmark=stereo&result=226494ba5559e9f5f46bdbd681d1564fee78409e). I am experimenting with the first author of the hitnet paper to check the potential problems...

@Magicboomliu you can watch this code repository to get timely code updates.

As my personal computing resources are relatively tight at the moment, people with adequate gpu computing resources are also welcome to contact me. email: [email protected] wechat: 16600286930

For an experiment, generally 2 to 4 GPUs are needed. (2 is ok)

@DrugRui Sorry for the late reply my idea Thanks

结构光双目相机能够一定程度上克服弱纹理的匹配问题，后续几个月会开放相关数据 模型和效果图，敬请关注 Structural-light binocular camera can overcome the problem of weak texture matching to some extent. Relevant data models and renderings will be opened in the next few months. Please...

I will update the model and the pretrained model later.

@HaixiongLi I update the model and pretrain weight. Enjoy yourself!

@saulzar @tolgabirdal I update the code now. Enjoy yourself!

I will notice you later. Sorry for the late reply.