Chore/ablations#14
Merged
Merged
Conversation
Introduce EndToEndTrainer that simulates the inference loop during training to address exposure bias — the model learns on its own (possibly wrong) association histories instead of ground-truth ones. Key changes: - Add TDLPModel abstract base class with prepare_loss_inputs() and compute_cost_matrix() to replace isinstance checks in the trainer - Add EndToEndTrainer with autoregressive forward pass, Hungarian matching, selective gradient frames, and track state management - Add trainer factory and TrainerType enum for config-driven selection - Add exp05, E2E training config, model config, and augmentation config Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Wraps the TDLP tracker as a Motrack-registered tracker and the DanceTrack
dataset as a Motrack-registered dataset, so motrack's run_inference,
run_eval, and run_optimize pipelines drive TDLP end-to-end. Adds
tools/motrack_{inference,eval,optimize}.py and the configs/motrack/
hydra tree. Bumps motrack to 0.7.0[mlflow] and wires MLFLOW_TRACKING_URI
through docker-compose on the shared motrack-net network.
Verified: HOTA=61.86 on DanceTrack val with the exp108b checkpoint,
matching the historical 61-62 range.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds 17 DanceTrack ablation configs under history/DanceTrack/ablations/
covering the four ablations the reviewers asked for:
- f_inter (object interaction encoder) 2x2 design:
tdlp_bboxes_NoInter, tdlp_NoInter, tdlp_FullNoInter, tdlp_MMOnlyInter,
tdlp_keypoints_NoInter, tdlp_appearance_NoInter
- link prediction head: tdlp_bboxes_Head{Concat,Diff,Dot}
- positive-weight (w+): tdlp_bboxes_PosWeight{1,20,100}
- multi-modal fusion: tdlp_Agg{LinearSum,Concat,Attn,Query}
+ tdlp_AggLinearSumRerun (seed-variance check)
Supporting code:
- TDCPConcatMLPAggregator added to aggregators.py (concat -> MLP fusion).
- TDSPConcatMLPHead and TDSPDotProductHead added to similarity_prediction.py.
- Fix _convert_data in tracker/online.py to populate all configured features
(was only writing bbox -- caused MM modes to silently receive zero
tensors for keypoints and appearance, ~4 HOTA drop).
- docker-compose: add /mnt/data bind mount for the secondary dataset disk.
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
No description provided.