requirements.txt

# === Core Runtime ===
gradio>=3.50           # Web interface
torch>=2.0             # Deep learning base
torchvision>=0.15      # For DeepLabV3 and image utilities
ultralytics>=8.0.11       # YOLOv5/v8/11 model loader
opencv-python>=4.7     # Video/image input/output
numpy>=1.21            # Array math
Pillow>=9.0            # Image loading, drawing

# === Hugging Face Models ===
transformers>=4.30     # SegFormer + tokenizer
huggingface-hub>=0.13.3  # For model downloads via hf_hub_download
sentencepiece          # Required by some HF models and processors

# === Depth Estimation (MiDaS, DPT) ===
# MiDaS needs scipy and possibly timm depending on config
scipy>=1.7
timm>=0.6.12            # For DPT/ViT backbones (needed for some MiDaS/DPT HF models)

# === Utility and Execution ===
requests>=2.28         # For downloading remote media
timeout-decorator>=0.5.0  # For analysis timeout handling
tqdm>=4.64             # Progress bar for batch ops
pandas>=1.3            # Optional: JSON structuring / metrics