search

facebookresearch / habitat-challenge

Code for the habitat challenge
https://aihabitat.org
MIT License
309 stars 56 forks source link

DDPPO Objectnav Agent and pretrained model #43

Closed mathfac closed 4 years ago

mathfac commented 4 years ago

Motivation and Context

Provide agent wrapper, pre-trained model and submit instructions for DDPPO Objectnav baseline.

How Has This Been Tested

Objectnav local test: docker build . --file Objectnav_DDPPO_baseline.Dockerfile -t objectnav_submission;bash test_locally_objectnav_rgbd.sh

Pointnav local test: docker build . --file Pointnav.Dockerfile -t pointnav_submission; ./test_locally_pointnav_rgbd.sh --docker-name pointnav_submission

dhruvbatra commented 4 years ago

~I was unable to test creating the DDPPO_Objectnav docker because my GCP VM kept going unresponsive every time I would try to. I suspect it's either filling up memory or space. But no point keeping this PR waiting.~

Edit: Just had to create a bigger memory VM. Docker container was created just fine. However, now I see 0 SPL with local testing. Is this expected: 

Successfully built fdd805e9ae39
Successfully tagged objectnav_submission:latest
(habitat) dbatra@ubuntu-bionic-1:~/habitat-challenge$ ./test_locally_objectnav_rgbd.sh
Neither `ifconfig` (`ifconfig -a`) nor `ip` (`ip address show`) commands are available, listing network interfaces is likely to fail
2020-05-08 05:55:59,800 Overwriting CNN input size of depth: (256, 256)
2020-05-08 05:55:59,803 Overwriting CNN input size of rgb: (256, 256)
Checkpoint loaded: demo.ckpt.pth
2020-05-08 05:56:24,742 Initializing dataset ObjectNav-v1
2020-05-08 05:56:24,851 initializing sim Sim-v0
2020-05-08 05:56:27,983 Initializing task ObjectNav-v1

2020-05-08 06:07:45,183 distance_to_goal: 5.4542582359164955
2020-05-08 06:07:45,183 success: 0.0
2020-05-08 06:07:45,183 spl: 0.0
dhruvbatra commented 4 years ago

@Skylion007 -- the DD-PPO PointNav docker creations is failing: 

(habitat) dbatra@ubuntu-bionic-1:~/habitat-challenge$ ls
CODE_OF_CONDUCT.md                   configs
CONTRIBUTING.md                      ddppo_agents.py
LICENSE                              ddppo_objectnav_habitat2020_challenge_baseline_v1.pth
Objectnav.Dockerfile                 habitat-challenge-data
Objectnav_DDPPO_baseline.Dockerfile  res
Pointnav.Dockerfile                  submission.sh
Pointnav_DDPPO_baseline.Dockerfile   test_locally_objectnav_rgbd.sh
README.md                            test_locally_pointnav_rgbd.sh
agent.py

(habitat) dbatra@ubuntu-bionic-1:~/habitat-challenge$ docker build . --file Pointnav_DDPPO_baseline.Dockerfile -t pointnav_submission
Sending build context to Docker daemon  98.69MB
Step 1/11 : FROM fairembodied/habitat-challenge:testing_2020_habitat_base_docker
 ---> bf52971c4a7f
Step 2/11 : RUN /bin/bash -c ". activate habitat; pip install ifcfg tensorboard && pip install torch==1.5.0+cu101 torchvision==0.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html"
 ---> Using cache
 ---> afbd1151493e
Step 3/11 : RUN /bin/bash -c ". activate habitat; git clone http://github.com/facebookresearch/habitat-api.git habitat-api2 && (cd habitat-api2 && git checkout 959bd45431edd8024832a877bdc8218015d97a7e) && cp -r habitat-api2/habitat_baselines habitat-api/."
 ---> Using cache
 ---> 20549b03767e
Step 4/11 : ADD ddppo_agents.py agent.py
 ---> Using cache
 ---> 70356dcb1fc5
Step 5/11 : ADD submission.sh submission.sh
 ---> Using cache
 ---> e854442eb860
Step 6/11 : ADD configs/challenge_pointnav2020.local.rgbd.yaml /challenge_pointnav2020.local.rgbd.yaml
 ---> Using cache
 ---> 70bd0a1adcd2
Step 7/11 : ADD configs/ configs/
 ---> Using cache
 ---> 4e2a35aa0b2f
Step 8/11 : ADD ddppo_pointnav_habitat2020_challenge_baseline_v1.pth demo.ckpt.pth
ADD failed: stat /var/lib/docker/tmp/docker-builder630335240/ddppo_pointnav_habitat2020_challenge_baseline_v1.pth: no such file or directory
mathfac commented 4 years ago

@dhruvbatra you missed wget https://dl.fbaipublicfiles.com/habitat/data/baselines/v1/ddppo_pointnav_habitat2020_challenge_baseline_v1.pth, double checking if it's still there in the doc.