Spaces:
Sleeping
Sleeping
Commit History
-
test: Optimize test_batch_inference to actually perform concurrent stress testing using asyncio.gather
Ha Trong Nguyenc7c0033 -
chore: Sync all latest updates
Ha Trong Nguyenc539a75 -
fix: Add httpx retries to prevent RemoteProtocolError from flaky HCMC server
Ha Trong Nguyen14dfbb8 -
fix: Add opencv-python-headless to requirements.txt for heatmap generation
Ha Trong Nguyen7bb0212 -
feat: Add OpenCV-based heatmap_base64 generation endpoint
Ha Trong Nguyen775374a -
feat: implement image fallback cache to handle intermittent camera API disconnects gracefully
Ha Trong Nguyenae4a838 -
feat: release final optimized ONNX 320x320 pipeline
Ha Trong Nguyen03eb31b -
feat: add batch inference testing script and update documentation
Ha Trong Nguyenefe5fd6 -
fix: resolve ModuleNotFoundError for models import
Ha Trong Nguyen- Move 'from models import get_model' inside load_model() so it executes AFTER sys.path is updated with ZIP_PROJECT_ROOTe800115 -
perf: optimize cpu inference speed
Ha Trong Nguyen- Reduce input size from 448 to 320 to match model training and reduce FLOPs - Set torch.set_num_threads(2) to prevent thread thrashing on HF free tier0255d68 -
fix: resolve CORS and swagger docs issues
Ha Trong Nguyen- Update CORS to allow all origins ('*') for frontend compatibility - Force enable Swagger UI and ReDoc endpointscd9be41 -
chore: add .gitignore + update ZIP model checkpoint
Ha Trong Nguyen- Add comprehensive .gitignore (Python cache, venvs, IDE, logs) - Update best_mae_0.pth with latest 2-class training checkpoint - Checkpoint tracked via Git LFS (.gitattributes)7e74d3a
-
perf: add 120s TTL cache for batch_predict to prevent CPU overload and reduce batch max to 15
Ha Trong Nguyenc1c828e -
feat: upgrade ZIP prediction model, add batch_predict API, caching and concurrency semaphores
Ha Trong Nguyen25d2309
-
Fix multi-class output: read per-class density maps (ch0=car, ch1=motorbike) instead of heuristic 70/30 split
Ha Trong Nguyen- model_service.py: detect 2-channel tensor output (B,2,H,W) and extract car_count from channel 0, motorbike_count from channel 1 directly - Heuristic 70/30 split now only used as fallback for single-class models - schemas.py: add car_count and motorbike_count fields to PredictionResult - config.py: add vehicle_split_motorbike_ratio for fallback config - Remove unused copy_script.pya69c54d
-
Fix ZIP_PROJECT_ROOT path resolution for Docker structure
Ha Trong Nguyen8d65e65 -
Add required HF Space README config
Ha Trong Nguyen92284c9 -
Fix HF Space dependencies
Ha Trong Nguyen154800b -
Initial commit for HuggingFace Space backend
Ha Trong Nguyeneb8e8ab