perf: Prevent GPU OOM and Improve Processing Efficiency in estimate_pose module

[tankchenggeng]

📝 What does this PR do?

This PR introduces a major refactor of the estimate_ose module in video2motion.py to address critical memory and performance limitations:

1. Generator for Lazy Loading: Replaces the list used for loading images to the device (GPU) with a generator.
2. Chunk Processing: Implements a chunking mechanism to process data_chunks in smaller, manageable batches.

🚫 Motivation and Context

• GPU Memory Overflow (OOM): The original code loaded all video frames into GPU memory at once ([frame.to(device) for frame in all_frames]), causing CUDA Out Of Memory errors with longer videos.
• Exponential Time Complexity: The processing time of the original algorithm scaled poorly with input size, making long video processing times impractical.

These changes transform the resource usage from being a function of the total video length to a function of the processing batch size.

🔧 Changes Made

File Modified: video2motion.py

Specific Changes:

1. Generator Expression:

   • func(images_crop): Used a generator expression.

2. Chunk Processing Logic:

   • Refactored the processing loop to break down the input data_chunks into smaller, fixed-size batches.
   • This prevents the algorithm from processing excessively large amounts of data in a single operation, changing the time complexity from exponential to near-linear for the overall video processing task.

🧪 How Has This Been Tested?

1. Functional Regression Test:
   • Ran the script on a short video (<5s) and verified the output matches the original behavior.
2. Performance & Stress Test:
   • Processed a long video (>5 minutes) that previously caused an OOM error. The script now completes successfully with manageable memory usage and processing time.
   • Monitored GPU memory usage via nvidia-smi. The memory footprint is now stable and significantly lower throughout the entire process, instead of climbing until a crash.

✅ Checklist

• My code follows the established code style of this project.
• I have thoroughly tested my changes, both for correct functionality and performance improvements.
• This change is a refactor of existing code and does not require documentation updates.