Feat/fix face depth

[kunkunlin1221]

!!! Git lfs unstable in container

[Copilot]

Pull Request Overview

This PR implements face depth functionality and refactors the face detection testing infrastructure. The changes move from hardcoded test expectations to file-based test data storage and add comprehensive face depth testing capabilities.

Key changes:

• Replaces inline test data with NPZ file-based test expectations for better maintainability
• Adds complete face depth testing module with batch processing support
• Fixes batch size handling bug in multiple face processing components

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/test_face_detection.py	Refactored to use NPZ files for test data and added batch testing
tests/test_face_depth.py	New comprehensive test suite for face depth functionality
tests/resources/answer/*.npz	Git LFS tracked test data files for face detection and depth
pyface/utils.py	Removed deprecated backend detection utilities
pyface/components/face_recognition/arcface.py	Fixed batch size variable naming for clarity
pyface/components/face_gender.py	Fixed batch size variable naming for clarity
pyface/components/face_detection/scrfd.py	Fixed batch size variable naming for clarity
pyface/components/face_depth/tddfav2.py	Fixed batch processing, type hints, and data handling
.gitattributes	Added NPZ files to Git LFS tracking

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[kunkunlin1221]

This pull request introduces several improvements and refactoring across demo scripts, CI workflows, and core components. The main focus is on standardizing resource paths in demo scripts, enhancing the CI pipeline (especially for CUDA builds), fixing batch handling in model inference, and updating Dockerfile configurations for compatibility and maintainability.

Demo script improvements:

• Standardized resource directory usage in all demo scripts (demo/face_detection.py, demo/face_depth.py, demo/face_gender.py, demo/face_landmark.py, demo/face_normalization.py, demo/face_recognition.py, demo/face_service.py). This replaces previous ad-hoc path handling with a consistent RESOURCE_DIR definition pointing to test resources. Output directories are also consistently created using cb.get_curdir(__file__) / "output". [1] [2] [3] [4] [5] [6] [7]

CI/CD and Dockerfile enhancements:

• Added a dedicated Dockerfile for PR builds (docker/pr.dockerfile) using nvidia/cuda:12.8.1-cudnn-runtime-ubuntu24.04, and updated the main Dockerfile to match the new base image for improved compatibility. [1] [2]
• Refactored the CUDA CI workflow (.github/workflows/cuda-ci.yml) to build and use a custom Docker image, retrieve runner UID/GID, and run jobs inside containers with GPU support. This improves reproducibility and resource management for GPU-based CI.
• Minor improvements to the CPU CI workflow (.github/workflows/cpu-ci.yml), including renaming steps for consistency and clarity. [1] [2]

Core component bug fixes and refactoring:

• Fixed batch size handling in model inference calls for face detection, gender detection, and depth estimation components. This ensures correct splitting and concatenation of inference results, preventing shape mismatches and potential runtime errors. [1] [2] [3]
• Minor code style and type annotation improvements in pyface/components/face_depth/tddfav2.py, including function signature updates and pylint disables for import errors. [1] [2] [3] [4] [5]

Repository maintenance:

• Removed LFS filters for image files in .gitattributes, likely to simplify repository management and avoid LFS-related issues for .jpg and .png files.