Update requirements.txt

requirements.txt

@@ -1,41 +1,55 @@
 # ============================================
-# WORKING REQUIREMENTS FOR MAXIMUM RESOURCE USAGE
-# Fixed for HuggingFace Spaces deployment
-# Target: 32GB RAM + 24GB GPU VRAM
+# PRODUCTION REQUIREMENTS - FINAL VERSION
+# MatAnyone (BEST) + SAM2 (VIDEO) + Rembg (FAST)
+# Optimized for HuggingFace Spaces L4 GPU
 # ============================================
 
 # Core Dependencies
 streamlit==1.32.2
-opencv-python-headless==4.9.0.80
-numpy==1.24.3
+opencv-python-headless==4.8.1.78  # Fixed for compatibility
+opencv-contrib-python==4.8.1.78   # Matching version
+numpy==1.26.4
 Pillow==10.2.0
 
-# System Monitoring - ESSENTIAL
+# System Monitoring
 psutil==5.9.8
 GPUtil==1.4.0
 nvidia-ml-py==12.535.161
 py3nvml==0.2.7
 
-# PyTorch with CUDA (HuggingFace provides CUDA 11.8)
-torch==2.1.2
-torchvision==0.16.2
-torchaudio==2.1.2
+# PyTorch with CUDA (Latest stable)
+torch==2.2.0
+torchvision==0.17.0
+torchaudio==2.2.0
 
-# Transformers & HuggingFace
-transformers==4.37.2
-accelerate==0.26.1
-datasets==2.16.1
-tokenizers==0.15.1
-safetensors==0.4.2
-huggingface-hub==0.20.3
+# ============================================
+# PRIMARY METHOD: MatAnyone (Production Quality)
+# ============================================
+git+https://github.com/pq-yang/MatAnyone.git
 
-# Background Removal
+# ============================================
+# SECONDARY METHOD: SAM2 (Video Optimized)
+# ============================================
+git+https://github.com/facebookresearch/segment-anything-2.git
+
+# ============================================
+# TERTIARY METHOD: Rembg (Fast Fallback)
+# ============================================
 rembg==2.0.53
 onnxruntime-gpu==1.16.3
 
-segment-anything==1.0
+# ============================================
+# Supporting Libraries
+# ============================================
 
-# Computer Vision Models
+# Transformers & HuggingFace (for model downloads)
+transformers==4.37.2
+accelerate==0.26.1
+huggingface-hub==0.20.3
+safetensors==0.4.2
+tokenizers==0.15.1
+
+# Computer Vision
 timm==0.9.12
 albumentations==1.3.1
 ultralytics==8.1.18
@@ -49,9 +63,9 @@ imageio==2.33.1
 imageio-ffmpeg==0.4.9
 
 # Video Processing
-opencv-contrib-python==4.9.0.80
 moviepy==1.0.3
 av==11.0.0
+decord==0.6.0
 
 # ML Libraries
 xgboost==2.0.3
@@ -62,7 +76,16 @@ scikit-learn==1.4.0
 pandas==2.2.0
 pyarrow==15.0.0
 
-# Memory Profiling
+# Progress & UI
+tqdm==4.66.1
+rich==13.7.0
+
+# Web & Downloads
+requests==2.31.0
+httpx==0.26.0
+aiohttp==3.9.3
+
+# Memory Management
 memory-profiler==0.61.0
 pympler==1.0.1
 objgraph==3.6.1
@@ -72,16 +95,35 @@ joblib==1.3.2
 threadpoolctl==3.2.0
 
 # Utilities
-tqdm==4.66.1
-rich==13.7.0
 python-dotenv==1.0.1
+fire==0.5.0
+typer==0.9.0
+
+# ============================================
+# OPTIONAL: Additional Features
+# ============================================
 
-# Additional Vision Models (if space allows)
-segment-anything==1.0
+# Audio support (for videos with sound)
+librosa==0.10.1
+soundfile==0.12.1
+pydub==0.25.1
+
+# Advanced CV models
+open-clip-torch==2.24.0
+clip-interrogator==0.6.0
 groundingdino-py==0.4.0
 
-# Optional Heavy Libraries (comment out if build fails)
-# detectron2 - requires manual installation
-# tensorflow==2.15.0  # Another 2GB
-# jax==0.4.23  # Another GPU framework
-# kornia==0.7.1  # GPU accelerated vision
+# Monitoring
+wandb==0.16.3
+tensorboard==2.15.1
+
+# ============================================
+# NOTES:
+# 1. MatAnyone provides best quality with temporal consistency
+# 2. SAM2 is video-optimized segmentation
+# 3. Rembg is fast fallback
+# 4. OpenCV is always available (comes with opencv-python)
+# 5. Models download dynamically on first use
+# 6. Total package size: ~10-15GB
+# 7. Models add: 500MB-2GB depending on selection
+# ============================================