TianqiTang

Hi, thanks for your excellent work at first.There is one thing confuses me that inference time doesn't include interp layer only when testing.In the official code or paper, it looks...

Sorry for my unsuitable expression. I mean that the time may contain this line https://github.com/hellochick/ICNet-tensorflow/blob/4aeb4005a2c25636d4d9bb4914f1d641f591d7cf/evaluate.py#L100

Hi, I am interested in your new inference time.I have larger one than paper refers to due to GPUs.Would you mind updating it somewhere?Thanks a lot:) @hellochick

Hi, I find the slam-based agent may perform random action. The code can be viewed on https://github.com/facebookresearch/habitat-api/blob/71d409ab214a7814a9bd9b7e44fd25f57a0443ba/habitat_baselines/agents/slam_agents.py#L356 What is the purpose of random action? It may cause early stop in...

Hi, I meet the similar problem on Ubuntu 16.04.6 with two GeForce GTX 1080 Ti and one Tesla K40c. 1. I log in the server via ssh -X -Y name@ip...

I find similar inf issues in the Gibson dataset when I apply continuous action space rather than discrete action space. The initial agent tends to walk randomly, it seems easy...

HI, it looks like one bug in my code. I use ```quat_from_coeffs``` with the wrong input format ```[a, b, c, d]```, which leads to the wrong agent pose.

Here is another failure case, where ```geodesic_distance``` return inf. I reset the start position and rotation in this episode, which are the same position I find inf. It looks like...

Thanks a lot. So we can't know whether if the agent is in a different place before the geodesic distance returns inf. The solution is to go back to the...

Hi, I have the same issue, too. It seems that the progress is stuck when it tries to unpack "hm3d-minival-habitat.tar", and I have applied for the access to HM3D dataset....