Xin Wang
|
9bc89b556f
|
Add shared-AV multiwindow Qwen feature extraction
|
2026-09-17 03:00:05 +08:00 |
|
Xin Wang
|
a9f748d6a6
|
Add Qwen tail and chunk ablations
|
2026-09-17 02:32:08 +08:00 |
|
Xin Wang
|
655d27a4ca
|
Add ROMA frozen backbone extraction support
|
2026-08-27 13:35:59 +08:00 |
|
Xin Wang
|
f35b9108ac
|
Add overlap-aware turn prediction objectives
|
2026-08-26 15:03:56 +08:00 |
|
Xin Wang
|
43e0e14cf5
|
Add learnable last-four-layer query fusion
|
2026-08-26 10:29:43 +08:00 |
|
Xin Wang
|
dbef1fc173
|
Add final-layer mixture readout
|
2026-08-26 09:56:54 +08:00 |
|
Xin Wang
|
f6dbce30e4
|
Reuse global history frames in overlap working memory
|
2026-08-24 11:17:39 +08:00 |
|
Xin Wang
|
1731f2b6f7
|
Optimize overlap working-memory audio preprocessing
|
2026-08-24 10:18:55 +08:00 |
|
Xin Wang
|
20446c0bd8
|
Add overlap working-memory ablation and LoRA epoch2 record
|
2026-08-24 09:52:56 +08:00 |
|
Xin Wang
|
c844673984
|
Add TorchCodec video backend
|
2026-08-23 16:41:05 +08:00 |
|
Xin Wang
|
5fa61eed23
|
Add extract_ego4d_batch.py for efficient EgoSpeak clip extraction. This script loads the Qwen model once and processes selected manifest rows sequentially, ensuring each clip is handled independently. Update README with usage instructions and details on the extraction process, including handling of completed outputs and memory management.
|
2026-08-18 22:22:37 +08:00 |
|
Xin Wang
|
43f47e8f3a
|
Add caching mechanism for completed AV blocks in streaming model. Introduce CompletedBlockPieceCache class to manage reusable blocks during strict-window replay, enhancing efficiency. Update profiling hooks for cache operations and refine README documentation to reflect changes in block handling and replay strategies.
|
2026-08-18 11:02:42 +08:00 |
|
Xin Wang
|
d610305355
|
Add profiling script for GPU bottleneck analysis in EasyCom extraction. Introduce command-line options for GPU ID, log directory, and sampling period. Implement stage profiling to measure extraction performance and synchronize CUDA operations. Enhance CausalSourceAV class with new parameters for decoder threads and open readers, optimizing video decoding process.
|
2026-08-18 10:30:19 +08:00 |
|
Xin Wang
|
24cb92f1f4
|
Add support for NVDEC backend in video processing and enhance README documentation. Introduce a new script for extracting EasyCom sessions on specific GPUs, detailing session management and error handling. Update requirements.txt to include PyNvVideoCodec as an optional dependency. Refactor streaming_turn_model_directly.py to accommodate multiple decoder backends and improve frame indexing with ffprobe.
|
2026-08-18 08:41:54 +08:00 |
|
Xin Wang
|
de29e75734
|
Add scripts for preparing and extracting 5 FPS audio-video clips, including prepare_ego4d_2fps.py, run_extract_5fps_2pergpu.sh, and run_prepare_all_5fps.sh. Introduce streaming_turn_model_directly.py for direct extraction from original media. Update README.md to reflect changes in Qwen's streaming design and causal extraction modes. Add new AGENTS.md file for documentation.
|
2026-08-17 15:17:11 +08:00 |
|