Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

Added goal sensor argument back, hallenge_pointnav2020_v2, added passing ResnetPolicy params, added SOFT_SPL #47

Merged
merged 6 commits into from
May 12, 2020

Conversation

mathfac
Copy link
Contributor

@mathfac mathfac commented May 11, 2020

Motivation and Context

  • Added goal sensor argument back to fix failure of PointNav docker
  • Added challenge_pointnav2020_v2 with camera tilt -20 degrees to comply with Gibson 2020 challenge, the config won't be used for Habitat challenge evaluation.
  • Expanded random seed setting to replicate same metrics with multiple runs.
  • Added passing ResnetPolicy params to address PPO model training with habitat 2020 challenge config #26 (comment)
  • Added SOFT_SPL metric to be consistent with metrics we report remotely.

How Has This Been Tested

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

2020-05-11 20:07:47,179 Initializing dataset ObjectNav-v1
2020-05-11 20:07:47,240 initializing sim Sim-v0
2020-05-11 20:07:48,833 Initializing task ObjectNav-v1
2020-05-11 20:12:07,349 distance_to_goal: 5.4542582359164955
2020-05-11 20:12:07,349 success: 0.0
2020-05-11 20:12:07,349 spl: 0.0
2020-05-11 20:12:07,349 softspl: 0.07157614435882213

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

2020-05-11 18:58:25,851 Overwriting CNN input size of depth: (256, 256)
2020-05-11 18:58:25,853 Overwriting CNN input size of rgb: (256, 256)
Checkpoint loaded: demo.ckpt.pth
2020-05-11 18:58:28,118 Initializing dataset PointNav-v1
2020-05-11 18:58:28,119 initializing sim Sim-v0
2020-05-11 18:58:29,471 Initializing task Nav-v0
2020-05-11 19:01:58,190 distance_to_goal: 2.2835945556561152
2020-05-11 19:01:58,190 success: 0.06666666666666667
2020-05-11 19:01:58,190 spl: 0.027204928901281608
2020-05-11 19:01:58,190 softspl: 0.2964175495801797

@facebook-github-bot facebook-github-bot added the CLA Signed Do not delete this pull request or issue due to inactivity. label May 11, 2020
@mathfac mathfac requested a review from abhiskk May 11, 2020 18:41
ddppo_agents.py Outdated Show resolved Hide resolved
ddppo_agents.py Outdated
normalize_visual_inputs="rgb" if config.INPUT_TYPE in ["rgb", "rgbd"] else False,
)
if "ObjectNav" not in config.TASK_CONFIG.TASK.TYPE:
policy_arguments[goal_sensor_uuid] = "pointgoal"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I why should this be hard coded instead of being read from the TASK_CONFIG though as per the earlier comment though?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@Skylion007 it was updated.

@mathfac mathfac changed the title Added goal sensor argument back, added passing ResnetPolicy params, added SOFT_SPL Added goal sensor argument back, hallenge_pointnav2020_v2, added passing ResnetPolicy params, added SOFT_SPL May 12, 2020
@mathfac mathfac merged commit 9810ccf into master May 12, 2020
@mathfac mathfac deleted the fix_goal_sensor branch May 12, 2020 00:51
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
CLA Signed Do not delete this pull request or issue due to inactivity.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants