Add GPU scheduling for benchmark evaluation scenes#4286
Conversation
Summary of ChangesHello, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request significantly enhances the benchmark evaluation process by enabling efficient multi-GPU utilization for reconstruction scenes. It introduces a flexible mechanism for GPU selection and task distribution, allowing for faster and more scalable evaluations. Additionally, it includes a minor but important update to ensure compatibility with newer NumPy versions. Highlights
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here. Footnotes
|
There was a problem hiding this comment.
Code Review
This pull request introduces GPU scheduling for benchmark evaluation, which is a valuable addition for multi-GPU setups. The implementation uses a round-robin approach to distribute scenes across specified or auto-detected GPUs. The code is generally well-structured. I've provided a couple of suggestions to improve the robustness of the GPU index parsing and to encourage adding tests for the new logic. The fix for the NumPy deprecation is also correct.
- Replace --parallelism with --num_threads and --num_parallel_scenes for finer control over total threads vs parallel scene count. - Capture per-step COLMAP output to extraction.log, matching.log, reconstruction.log, alignment.log in each scene workspace so parallel runs remain debuggable. - Make Ctrl+C cleanly terminate workers and child COLMAP processes via PR_SET_PDEATHSIG and a SIGINT-ignoring pool initializer.
2b21568 to
e4b5127
Compare
Distribute reconstruction scenes across available GPUs using round-robin scheduling. Add --gpu_index argument to control which GPUs to use, with auto-detection of all CUDA devices by default. Each scene is assigned a specific GPU index, enabling efficient multi-GPU utilization during parallel benchmark evaluation.
e4b5127 to
8cfda08
Compare
Distribute reconstruction scenes across available GPUs using round-robin scheduling. Add --gpu_index argument to control which GPUs to use, with auto-detection of all CUDA devices by default. Each scene is assigned a specific GPU index, enabling efficient multi-GPU utilization during parallel benchmark evaluation.
Also fix numpy deprecation: np.acos -> np.arccos.