Adding docker support

[ShubhamSureshPatil]

This PR adds comprehensive Docker support to UltraSam, providing a containerized environment that resolves dependency conflicts and simplifies setup across different systems.

Changes
Dockerfile: Complete containerized environment with PyTorch 2.0.0, CUDA 11.7, and pre-compiled MMCV extensions with MMlab libraries installed.

Key Features
🐳 One-command setup with all dependencies.
🚀 GPU acceleration: Full NVIDIA Container Toolkit support
📦 Dependency resolution: All OpenMMLab suite with compatible versions (NumPy 1.26.4 for tensor compatibility)
📝 Comprehensive docs: Quick start guide with Docker alternatives

Benefits

• Eliminates MMCV compilation issues across different CUDA environments
• Resolves PyTorch/NumPy version conflicts automatically
• Provides consistent environment for reproducible research
• Simplifies onboarding for new contributors
• Supports both inference and training workflows

Testing
✅ Sample inference on provided MMOTU2D dataset
✅ Training pipeline verification with UltraSAM_point_refine configuration
✅ GPU acceleration confirmed on multiple NVIDIA GPUs
✅ Cross-platform compatibility (Linux with NVIDIA drivers)

[ShubhamSureshPatil]

MMlab, MMCV support with latest CUDA (12 and above) is dicey. Highly recommend docker support.

[Chop1]

Hi, thanks for your pr.

Could you update the readme to have the whole docker info in a separated section, between "Minimal working example" and "usage"?
Which would be folded as in the main branch.

Thanks!

[ShubhamSureshPatil]

Thank you for the feedback!

Changes have been made as requested.

I also added a visual_inference.py script to quickly test on the sample dataset provided as part of the repo using docker.

Any feedback is appreciated. Thanks!

[Chop1]

Why do we need the visual_inference.py? Cant we run the same test command using mim inside the docker?