hexone2086 56aa10a503 feat: per-step supervision over full sequence ...

- model outputs (B, S, 2) instead of (B, 2) — GRU output at every timestep
- train/val loss computed over all S timesteps with reduction=none
- benchmark/evaluate.py takes pred[:, -1, :] for final-step evaluation
- added per-step loss logging (8 evenly spaced steps) to TensorBoard

Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>

2026-07-09 15:49:23 +08:00

benchmark

feat: per-step supervision over full sequence

2026-07-09 15:49:23 +08:00

src

feat: per-step supervision over full sequence

2026-07-09 15:49:23 +08:00

visualize

feat: add --show-events overlay with raw log intensity

2026-06-08 11:47:19 +08:00

.gitignore

initial commit

2026-05-29 18:49:01 +08:00

AGENTS.md

feat: replace non-overlapping windows with sliding-window sequence sampling

2026-06-05 16:45:24 +08:00

batch_convert.sh

initial commit

2026-05-29 18:49:01 +08:00

DATASET_FORMAT.md

initial commit

2026-05-29 18:49:01 +08:00

download_davis_gt_rosbags.sh

initial commit

2026-05-29 18:49:01 +08:00

requirements.txt

initial commit

2026-05-29 18:49:01 +08:00

rosbag2wds.py

initial commit

2026-05-29 18:49:01 +08:00

start_ros_container.sh

fix: make start_ros_container.sh executable

2026-06-04 20:37:58 +08:00

S

Description

No description provided

112 KiB

Languages

Python 90.2%

Shell 9.8%