Spaces:
Sleeping
Sleeping
Commit History
-
fix: eye gaze toggle now properly controls gaze line visibility
k230994623dc18ac -
fix: add libgles2 libegl1 for MediaPipe OpenGL ES dependency
k22056537All pipelines were failing with: libGLESv2.so.2: cannot open shared object file0d448a6 -
fix: use huggingface_hub to resolve LFS pointers at runtime
k22056537urllib can't follow HF's Xet CDN redirects for LFS files. Switched to hf_hub_download which handles auth and storage backends properly.44572a8 -
fix: resolve LFS joblib/npz files at runtime before app starts
k22056537Small LFS-tracked files (joblib, npz) are pointers in Docker. start.sh downloads real files from HF API before uvicorn launches so model pipelines can load them.6dcb046 -
fix: match integration_test2 Dockerfile and config
k22056537Replicate the working setup from FocusGuard/integration_test2: - Same Dockerfile structure with npm build and model downloads - Same .dockerignore (excludes .git, dev files) - Same .gitattributes (only pkl/npz/joblib as LFS) - Un-LFS .pt files so they're real objects in Docker contextc2bffa9 -
fix: resolve LFS at build time via git lfs pull
k22056537Include .git in Docker context so git lfs pull can convert pointers to real files. Restored model download steps from original Dockerfile.241eb23 -
fix: add dockerignore, restore git dep, trim requirements
k22056537- Added .dockerignore to exclude .git (516MB), node_modules, dist - Restored git in apt (needed for pip face_detection package) - Removed torch/torchvision from requirements.txt (already installed separately with CPU index to avoid pulling 2.5GB CUDA build) - Dropped clearml, pytest from prod requirements8eeba10 -
fix: ship pre-built frontend, drop npm from Docker build
k22056537Avoids Node.js/npm version issues and OOM in HF build env. Frontend is built locally and committed as static/.ef3b36a -
fix: resolve LFS pointers at runtime instead of build time
k22056537Build-time downloads fail (circular dep — files not served while building). Moved LFS resolution to a startup script that runs before uvicorn launches.95de31d -
fix: make face_mesh model pre-download non-fatal during build
k22056537Network restrictions in HF build env can block Google Storage. The app will retry the download at startup if it fails here.d14f41e -
fix: download LFS files from HF API during Docker build
k22056537HF Docker builds copy LFS pointers, not real files. Added a build step that detects pointer files and downloads the actual binaries from the HF Hub resolve API.d5c34f9 -
fix: restore LFS for all binary files, add git-lfs to Docker build
k22056537HF rejects any binary files not tracked by LFS. Re-added LFS patterns for *.npz, *.pt, *.joblib, etc. Added git-lfs install to Dockerfile with a fallback to pull real files if pointers are copied, plus build-time checkpoint verification.765fbe6 -
fix: remove aggressive LFS patterns, keep only L2CS weights as LFS
k22056537Small checkpoints (mlp_best.pt, scaler_mlp.joblib, etc.) were being stored as LFS pointers which broke model loading on HF Spaces. Only the 91 MB L2CSNet_gaze360.pkl needs LFS. Added build-time checkpoint verification to catch this earlier.739eedb -
chore: add HF Space docker metadata
k22056537Make the Space auto-detect the Docker runtime and port without extra manual config.0da4f80
-
feat: UI nav, onboarding, L2CS weights path + torch.load; trim dev files
k22056537- Home nav: drop avatar, add Home; simplify home CTAs - Focus onboarding + pipeline/L2CS fixes; history clear in Help only - Ignore coverage; Docker log info; remove dead FocusPage, test_data, fetch_sweepa75bb5a
-
Integrate L2CS-Net gaze estimation
Abdelrahman Almatrooshi- Add L2CS-Net in-tree (models/L2CS-Net/) with Gaze360 weights via Git LFS - L2CSPipeline: ResNet50 gaze + MediaPipe head pose, roll de-rotation, cosine scoring - 9-point polynomial gaze calibration with bias correction and IQR outlier filtering - Gaze-eye fusion: calibrated screen coords + EAR for focus detection - L2CS Boost mode: runs gaze alongside any base model (35/65 weight, veto at 0.38) - Calibration UI: fullscreen overlay, auto-advance, progress ring - Frontend: GAZE toggle, Calibrate button, gaze pointer dot on canvas - Bumped capture resolution to 640x480 @ JPEG 0.75 - Dockerfile: added git, CPU-only torch for HF Space deploymente557410 -
change test params and files
a56564c -
Merge feature/clearml-thresholds
982620c -
Sorry the previous commit came from feature/ui-fix not feature/integration-cleaned xD
3d761f3 -
MERGE EVERYTHING from feature/integration-cleaned :((((((((((((((((((((((((((((((((((((((((((((((( so tired
2be007a -
Merge prepare_dataset.py from feature/ui-fix
fd607ef -
Merge all other things coming from feature/integration
2aa12be -
Merge testing scripts from feauture/intergration
790e2c2 -
Start merging everything before 17/03/26
6d2f7ec
-
evaluation: channel ablation script + feature importance LOPO
k22056537e69e3a3 -
refactor: single checkpoints/ dir, fix uvicorn cmd and hybrid config path
k22056537266aee6
-
chore: MLP pipeline, evaluation updates, feature importance, confusion matrices
k220565378b47064 -
chore: integration cleanup — remove eye CNN, add threshold justification, fix pipeline
k220565376114098
-
add api test
633f159 -
Merge branch 'feature/integration' of https://github.kcl.ac.uk/k23172173/GAP_Large_project into feature/integration
598a3d8 -
Update .gitignore
175b15e -
update params
0ab8f3d -
Combine all updated main code together, forming new main.py
9429ca0 -
Fix open-cv GPU support problem (opencv library changed)
bf0bdbc -
Combined main.py and main_yolo.py to create main_combined.py, where main.py came from Mohamad and main_yolo.py came from UI team. (To be tested)
357e9dc -
Merge UI Manually (expect main)
dadceb2 -
Put all the models together (expect UI)
4a5bfab
-
integrated open/closed eye cnn in script
Abdelrahman Almatrooshi82d2ab7
-
Merge pull request #7 from k23172173/dev
Yingtao ZhengUnverified signatureMerge pull request #6 from k23172173/main2eaf50d -
Merge pull request #6 from k23172173/main
Yingtao Zheng (k23158987)Unverified signatureUpdate Dev to newest version, correct wrong branch mistakes52d831a
-
Add Ayten attention dataset session 1
Ayten Arab2ea6266
-
Replace CNN_MODEL submodule with actual files
Saba Al-Gafrif001a13 -
Add CNN model
Saba Al-Gafri15adefa -
Add Saba collected focus dataset
Saba Al-Gafri0c5d783 -
Merge pull request #5 from k23172173/feat/facemesh
Yingtao Zheng (k23158987)Unverified signatureMerging for Updated code97f5eb9 -
Merge pull request #4 from k23172173/main
Yingtao Zheng (k23158987)Unverified signatureAdd Abdelrahman's data and Kexin's second data95d9f98 -
Data Collection Yingtao
Yingtao Zheng (k23158987)Unverified signatureccdf470 -
Merge branch 'main' of https://github.kcl.ac.uk/k23172173/GAP_Large_project
ac7d170
-
Abdelrahman's data
Abdelrahman Almatrooshi9e9c88a -
Add kexin's second record data file
Kexin Wang (k23168350)75b5135