In TensorRT 10.x, INT8 conversion requires FP16 to be set as a fallback for
network layers (e.g. normalization ops in detection models) that have no INT8
kernel implementation. Without FP16, build_serialized_network can return None
on Jetson for YOLO-type models. INT8 flag is still the primary precision;
FP16 is only the layer-level fallback within the same engine.
Made-with: Cursor
Tests expecting file storage (image/video write) failed because JWT_SECRET was
not set in the test environment, causing require_auth to return "" (falsy),
skipping the storage block. Both tests now patch main.JWT_SECRET and use a
properly signed JWT.
Made-with: Cursor
- Dockerfile.jetson: JetPack 6.x L4T base image (aarch64), TensorRT and PyCUDA from apt
- requirements-jetson.txt: derived from requirements.txt, no pip tensorrt/pycuda
- docker-compose.jetson.yml: runtime: nvidia for NVIDIA Container Runtime
- tensorrt_engine.pyx: convert_from_source accepts optional calib_cache_path; INT8 used when cache present, FP16 fallback; get_engine_filename encodes precision suffix to avoid engine cache confusion
- inference.pyx: init_ai tries INT8 engine then FP16 on lookup; downloads calibration cache before conversion thread; passes cache path through to convert_from_source
- constants_inf: add INT8_CALIB_CACHE_FILE constant
- Unit tests for AC-3 (INT8 flag set when cache provided) and AC-4 (FP16 when no cache)
Made-with: Cursor
- Pin all deps; h11==0.16.0 (CVE-2025-43859), python-multipart>=1.3.1 (CVE-2026-28356), PyJWT==2.12.1
- Add HMAC JWT verification (require_auth FastAPI dependency, JWT_SECRET-gated)
- Fix TokenManager._refresh() to use ADMIN_API_URL instead of ANNOTATIONS_URL
- Rename POST /detect → POST /detect/image (image-only, rejects video files)
- Replace global SSE stream with per-job SSE: GET /detect/{media_id} with event replay buffer
- Apply require_auth to all 4 protected endpoints
- Fix on_annotation/on_status closure to use mutable current_id for correct post-upload event routing
- Add non-root appuser to Dockerfile and Dockerfile.gpu
- Add JWT_SECRET to e2e/docker-compose.test.yml and run-tests.sh
- Update all e2e tests and unit tests for new endpoints and HMAC token signing
- 64/64 tests pass
Made-with: Cursor
- Add tests/test_az178_realvideo_streaming.py: integration test that validates
frame decoding begins while upload is still in progress using a real video fixture
- Add conftest.py: pytest plugin for per-test duration reporting
- Update e2e tests (async_sse, performance, security, streaming_video_upload, video)
and run-tests.sh for updated test suite
- Move AZ-178 task to done/; add data/ to .gitignore (StreamingBuffer temp files)
- Update autopilot state to step 12 (Security Audit) for new feature cycle
Made-with: Cursor
- Added `/detect/video` endpoint for true streaming video detection, allowing inference to start as upload bytes arrive.
- Introduced `run_detect_video_stream` method in the inference module to handle video processing from a file-like object.
- Updated media hashing to include a new function for computing hashes directly from files with minimal I/O.
- Enhanced documentation to reflect changes in video processing and API behavior.
Made-with: Cursor
Oleksandr Bezdieniezhnykh