IliasHad/edit-mind v0.14.1

on GitHub

Overview

This release brings significant improvements to video processing reliability, testing infrastructure, and UI improvements.

New Features

1. Added Docker image with GPU support via CUDA for the ML service
2. Added support to FFmpeg to utilize the GPU via CUDA for the background-jobs service
3. Added environment MAX_CONCURRENT_TRANSCRIPTIONS and MAX_CONCURRENT_ANALYSES variables to control how many video transcription and frame analysis processing jobs run in parallel

Bug Fixes

1. Add app version UI to the web UI and checker using the GitHub release
2. Immich importer update button issue
3. Remove the failed video processing jobs using the BullMQ
4. Update folderId when the folder is 2 levels deep or more for video processing jobs

Notes:

The GPU-enabled ML service image is ~10.7 GB (vs 3.6 GB CPU-only) due to unavoidable dependencies:

1. TensorFlow (1.8 GB) - Required by DeepFace for face recognition (when this issue is solved, we can work on using PyTorch as a backend for face recognition)
2. PyTorch + CUDA (1.5 GB) - Required by YOLO for object detection
3. NVIDIA CUDA Runtime (2.8 GB) - Base GPU libraries

Thank you @alikhanich for the help :)

Full Changelog: v0.14.0...v0.14.1