updating dockerfile and dependencies

CLAUDE.md

@@ -1,76 +0,0 @@
-# CLAUDE.md
-
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
-
-## Project Overview
-This is a OneFormer-based universal image segmentation application with integrated contrast detection capabilities. It supports panoptic, instance, and semantic segmentation using transformer-based models (Swin-L and DiNAT-L backbones).
-
-## Essential Commands
-
-### Setup and Installation
-```bash
-# Download model checkpoints from HuggingFace
-python setup.py --download
-
-# Build deformable attention CUDA operations (required)
-bash deform_setup.sh
-
-# Install NATTEN if needed
-cd NATTEN && pip install -e .
-```
-
-### Running the Application
-```bash
-# Main application launcher
-bash t.sh
-
-# Or run directly
-python gradio_test.py
-```
-
-### Testing
-```bash
-# Run NATTEN tests
-cd NATTEN && python -m unittest discover -v -s ./tests
-```
-
-## Architecture Overview
-
-### Core Components
-1. **OneFormer Model** (`oneformer/`): Universal segmentation transformer implementation
-   - `oneformer_model.py`: Main model class
-   - `modeling/`: Model components including backbones, decoders, and criterion
-   - `data/`: Dataset mappers and tokenizers for text queries
-
-2. **Gradio Applications** (multiple entry points):
-   - `gradio_test.py`: Main combined interface with pre/post-processing logic
-   - `gradio_app.py`: Basic segmentation interface
-   - `gradio_contrast.py`: Contrast detection focused interface
-
-3. **Contrast Detection** (`utils/`): Multiple contrast analysis algorithms
-   - `improved_contrast_analyzer.py`: Main analyzer combining all methods
-   - Individual analyzers: luminance, saturation, hue contrast detection
-
-4. **Model Configurations** (`configs/`): Pre-defined configs for:
-   - ADE20K (150 classes)
-   - Cityscapes (19 classes)
-   - COCO (133 classes)
-
-### Key Implementation Details
-- Models are loaded from HuggingFace hub (shi-labs organization)
-- Supports both CPU and GPU inference (auto-detection)
-- Custom CUDA kernels for multi-scale deformable attention
-- NATTEN library provides neighborhood attention operations
-
-### Model Checkpoints
-The application uses pre-trained models from:
-- `shi-labs/oneformer_cityscapes_swin_large`
-- `shi-labs/oneformer_coco_swin_large`
-- `shi-labs/oneformer_ade20k_swin_large`
-- DiNAT variants also available
-
-### Important Notes
-- Always run `deform_setup.sh` after environment setup for CUDA operations
-- The application expects model checkpoints to be downloaded before running
-- Gradio interface runs on port 7860 by default
-- GPU memory requirements vary by model (Swin-L models need ~12GB)

Dockerfile

@@ -23,16 +23,12 @@ RUN apt-get update && apt-get install -y \
 # Create user
 RUN useradd -m -u 1000 user
 
-# Install specific PyTorch CPU version FIRST
-RUN pip install torch==2.0.1+cpu torchvision==0.15.2+cpu torchaudio==2.0.2+cpu --index-url https://download.pytorch.org/whl/cpu --no-deps
-RUN pip install numpy pillow typing-extensions
+# Install PyTorch with ALL dependencies
+RUN pip install torch==2.0.1+cpu torchvision==0.15.2+cpu torchaudio==2.0.2+cpu \
+    --index-url https://download.pytorch.org/whl/cpu
 
-# Install detectron2
-RUN git clone https://github.com/facebookresearch/detectron2 /tmp/detectron2 && \
-    cd /tmp/detectron2 && \
-    git checkout v0.6 && \
-    pip install -e . --no-deps && \
-    pip install -r requirements.txt
+# Install detectron2 using pre-built wheel (much more reliable)
+RUN pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cpu/torch2.0/index.html
 
 # Switch to user
 USER user
@@ -45,13 +41,11 @@ ENV FORCE_CUDA="0"
 ENV TORCH_CUDA_ARCH_LIST=""
 
 # Copy and install requirements
+COPY --chown=user:user requirements.txt .
 RUN pip install --user --no-cache-dir -r requirements.txt
 
 # Copy app files
 COPY --chown=user:user . .
 
-# Install NATTEN last to ensure correct torch version
-RUN pip install --user natten==0.14.6+torch200cpu -f https://shi-labs.com/natten/wheels/cpu/torch2.0.0/index.html --no-deps
-
 EXPOSE 7860
 CMD ["python", "app.py"]

requirements.txt

@@ -1,8 +1,5 @@
-
-
-
-# Rest of your dependencies
-numpy>=1.23.0
+# Core dependencies (NO torch, NO detectron2, NO natten)
+numpy>=1.23.0,<2.0.0
 Pillow>=8.3.2
 opencv-python
 git+https://github.com/cocodataset/panopticapi.git
@@ -25,11 +22,10 @@ inflect
 diffdist
 gdown
 huggingface_hub
-pytorch_lightning
+pytorch_lightning==2.0.0
 tqdm
 gradio
 wget
 
-# NATTEN 
--f https://shi-labs.com/natten/wheels/cpu/torch2.0.0/index.html
-natten==0.14.6
+# NATTEN - install separately if needed
+# natten==0.14.6+torch200cpu -f https://shi-labs.com/natten/wheels/cpu/torch2.0.0/index.html