Refactor model configuration with a model registry #15
+145
−70
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add configs and model registry
… instead of model_path
…itialization to accept arguments.
…istry using get_model, replacing the previous model_path approach.
…e, and update saved_model_path for consistency with model registry usage.
…pe for model registry integration, and comment out model_path for clarity.
…ered model registry, enhancing integration with the model registry.
…_model_path for consistency with the new model registry structure.
…ing for checkpoints, including model and script files, improving data management during training.
… file reference for improved clarity and consistency.
deruyter92
force-pushed
the
ti_video_demo
branch
from
01eab70 to
80b2420
Compare
deruyter92
approved these changes
Feb 11, 2026
Collaborator
deruyter92
left a comment
deruyter92
left a comment
There was a problem hiding this comment.
Your changes look good to me!
A few remarks:
- please take a look at the configs to see if the defaults are correct
- are you happy with the model names? In case you want to change them (e.g. "fmpose3d" -> "fmpose3d_human"), now is still a good time, but later this will cause difficulties. All up to you what you think is best of course.
- Are the demo predictions for the animal model tracked on GitHub on purpose? If not, maybe better to delete them.
| return x | ||
|
|
||
| class Model(nn.Module): | ||
| @register_model("fmpose3d_animals") |
Collaborator
There was a problem hiding this comment.
Are you happy with the names "fmpose3d" for humans and "fmpose3d_animals" for animals? Or do you want to change "fmpose3d" -> "fmpose3d_humans" or something similar?
This would be a good moment to choose the final names, before these names will be used in other code as well.
Collaborator
Author
There was a problem hiding this comment.
Yes, I will change the name of the human model to 'fmpose3d_humans'
Comment on lines
28
to
47
| _FMPOSE3D_DEFAULTS: Dict[str, Dict] = { | ||
| "fmpose3d": { | ||
| "n_joints": 17, | ||
| "out_joints": 17, | ||
| "dataset": "h36m", | ||
| "sample_steps": 3, | ||
| "joints_left": [4, 5, 6, 11, 12, 13], | ||
| "joints_right": [1, 2, 3, 14, 15, 16], | ||
| "root_joint": 0, | ||
| }, | ||
| "fmpose3d_animals": { | ||
| "n_joints": 26, | ||
| "out_joints": 26, | ||
| "dataset": "animal3d", | ||
| "sample_steps": 3, | ||
| "joints_left": [0, 3, 5, 8, 10, 12, 14, 16, 20, 22], | ||
| "joints_right": [1, 4, 6, 9, 11, 13, 15, 17, 21, 23], | ||
| "root_joint": 7, | ||
| }, | ||
| } |
Collaborator
There was a problem hiding this comment.
@xiu-cs, please check if these default config values are correct and complete.
Collaborator
Author
There was a problem hiding this comment.
I have updated this; it looks great overall.
Comment on lines
+60
to
+64
| n_joints: int = INFER_FROM_MODEL_TYPE # type: ignore[assignment] | ||
| out_joints: int = INFER_FROM_MODEL_TYPE # type: ignore[assignment] | ||
| joints_left: List[int] = INFER_FROM_MODEL_TYPE # type: ignore[assignment] | ||
| joints_right: List[int] = INFER_FROM_MODEL_TYPE # type: ignore[assignment] | ||
| root_joint: int = INFER_FROM_MODEL_TYPE # type: ignore[assignment] |
Collaborator
There was a problem hiding this comment.
this is added now, so you can load an FMPose3DConfig from model_type with the appropriate number of joints etc:
model_cfg = FMPose3DConfig(model_type="fmpose3d_animal')
… files for consistency in the FMPose3D framework.
…onsistency across the FMPose3D framework.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This pull request updates the model loading and configuration system for FMPose3D and Animal3D workflows, moving from direct file path imports to a model registry approach. This makes it easier to specify models by name, improves reproducibility, and standardizes script arguments. It also updates pretrained model references and cleans up related scripts.
Model loading and registry improvements:
get_modeland--model_typefor both human and animal 3D pose estimation; updated code inmain_animal3d.py,vis_animals.py, andFMPose3D_main.pyto use this registry. [1] [2] [3]Model) with the registry via@register_model("fmpose3d_animals")and made it inherit fromBaseModel.--model_type(defaulting tofmpose3d_animals) and clarified the use of--model_pathas an override.Script and configuration updates:
train_animal3d.sh,test_animal3d.sh,vis_animals.sh,FMPose3D_train.sh,FMPose3D_test.sh) to use--model_typeinstead of--model_path, and set new default paths for pretrained weights. [1] [2] [3] [4] [5] [6] [7] [8] [9]Minor improvements and cleanup: