search

NVlabs / FoundationPose

[CVPR 2024 Highlight] FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects
https://nvlabs.github.io/FoundationPose/
Other
955 stars 99 forks source link

Hello, I encountered a model loading error while replacing the model from 2023-10-28-18-33-37 to 2024-01-11-20-02-45 #70

Closed wsq1010 closed 3 weeks ago

wsq1010 commented 4 weeks ago

self.run_name = "2024-01-11-20-02-45"

Traceback (most recent call last): File "run_demo_test.py", line 40, in refiner = PoseRefinePredictor() File "/home/bowen/FoundationPose/learning/training/predict_pose_refine.py", line 143, in init self.model.load_state_dict(ckpt) File "/opt/conda/envs/my/lib/python3.8/site-packages/torch/nn/modules/module.py", line 2041, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for RefineNet:

wenbowen123 commented 3 weeks ago

Hi, why do you want to change the model? They are for refine network and score network respectively. You cannot swap them.

wsq1010 commented 3 weeks ago

Hello, I see in run_demo that the external parameters of the camera are not used, only the internal parameters of the camera are used. Is there any problem with the coordinates in the camera coordinate system we obtained

wenbowen123 commented 3 weeks ago

Hello, I see in run_demo that the external parameters of the camera are not used, only the internal parameters of the camera are used. Is there any problem with the coordinates in the camera coordinate system we obtained

the pose you get is the object pose in camera coordinate. That's unrelated to extrinsic params (e.g. camera to world).

wsq1010 commented 3 weeks ago

Hello, is there any way to generate the required mask through obj and other models

wenbowen123 commented 3 weeks ago

This is a separate topic. let's discuss at https://github.com/NVlabs/FoundationPose/issues/88