Skip to content

update dependency version in README.md #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Apsinghsa wants to merge 2 commits into
base: main
Choose a base branch
from
Open

update dependency version in README.md #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
526b336
update dependency version in README.md
Apsinghsa Apr 23, 2026
2ff5850
fix install.yaml versions
Apsinghsa Apr 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
291 changes: 270 additions & 21 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,42 +1,291 @@
# FoundDiff
Official implementation of "FoundDiff: Foundational Diffusion Model for Generalizable Low-Dose CT Denoising"
# FoundDiff - Environment Setup Guide (Working)

## Approach
![](figs/network.png)
## Model Weights

Download the required model weights from: [Google Drive](https://drive.google.com/drive/folders/1B33XyPqC9KkmzmfrCq20-7Xxuf-23PMc?usp=sharing)

## Updates
Feb, 2026: Upload DA-CLIP and FoundDiff model weight (https://drive.google.com/drive/folders/1B33XyPqC9KkmzmfrCq20-7Xxuf-23PMc?usp=sharing)
July, 2025: initial commit.
You will need:
- `DA-CLIP.pth` - Place in `src/` (or as specified in DA-Diff.py)
- `model-400.pt` - Place in `checkpoints/FoundDiff/sample/`

## Quick Setup (Recommended)

## Data Preparation
The 2016 AAPM-Mayo dataset can be downloaded from: [CT Clinical Innovation Center](https://ctcicblog.mayo.edu/2016-low-dose-ct-grand-challenge/) (B30 kernel)
The 2020 AAPM-Mayo dataset can be downloaded from: [cancer imaging archive](https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=52758026)
Use the provided `install.yaml` to create the environment:

```bash
conda env create -f install.yaml
conda activate founddiff

## Requirements
# Install pre-built wheels for causal-conv1d and mamba-ssm
pip install causal-conv1d==1.2.2.post1 --no-build-isolation
pip install https://github.com/state-spaces/mamba/releases/download/v2.2.2/mamba_ssm-2.2.2+cu118torch2.1cxx11abiFalse-cp310-cp310-linux_x86_64.whl

# Verify
python -c "import selective_scan_cuda; print('CUDA kernels loaded successfully!')"
```

## System Requirements

- Linux Platform
- torch==1.12.1+cu113 # depends on the CUDA version of your machine
- torchvision==0.13.1+cu113
- Python==3.8.0
- numpy==1.22.3
- NVIDIA GPU with CUDA 11.8+ support
- Python 3.10

## Manual Step-by-Step Installation (Alternative)

### 1. Create Conda Environment

If you prefer to use the provided install.yaml, skip to step 4 after creating the environment:

```bash
conda create -n founddiff python=3.10 -y
conda activate founddiff
```

### 2. Install PyTorch 2.1.0 with CUDA 11.8

```bash
pip install torch==2.1.0 torchvision==0.16.0 --index-url https://download.pytorch.org/whl/cu118
pip install "numpy<2"
```

## Traning and & Inference
### 3. Install CUDA Toolkit 11.8

**Critical**: Use the specific nvidia channel label to get CUDA 11.8 (not newer versions):

#### Training:
```bash
conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit -y
```
CUDA_VISIBLE_DEVICES=1 python train.py --name FoundDiff --is_train --train_num_steps 400000

Verify installation:
```bash
nvcc --version
# Should show: CUDA compilation tools, release 11.8, V11.8.89
```

### 4. Install Prerequisites

```bash
pip install wheel packaging ninja
pip install causal-conv1d==1.2.2.post1 --no-build-isolation
```

### 5. Install mamba-ssm (Pre-built Wheel)

Download and install the pre-built wheel for CUDA 11.8 + Python 3.10:

```bash
pip install https://github.com/state-spaces/mamba/releases/download/v2.2.2/mamba_ssm-2.2.2+cu118torch2.1cxx11abiFalse-cp310-cp310-linux_x86_64.whl
```

### 6. Set Environment Variables

Create activation script to ensure torch libraries are found:

```bash
mkdir -p $CONDA_PREFIX/etc/conda/activate.d/
echo 'export LD_LIBRARY_PATH="$CONDA_PREFIX/lib/python3.10/site-packages/torch/lib:$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"' > $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export CUDA_HOME="$CONDA_PREFIX"' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export PATH="$CONDA_PREFIX/bin:$PATH"' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
```

#### Inference & testing:
Put DA-CLIP.pth in src/DA-Diff.py and model-400.pt in checkpoints/FoundDiff/sample
### 7. Verify Installation

```bash
conda activate founddiff
python -c "import selective_scan_cuda; print('CUDA kernels loaded successfully!')"
```

Should output: `CUDA kernels loaded successfully!`

## Additional Dependencies

The project requires several Python packages. Install them:

```bash
pip install kornia einops ipdb pywt accelerate ema-pytorch wandb Augmentor opencv-python lmdb scipy scikit-image matplotlib timm
```

Note: Some packages may require additional system dependencies or may have version conflicts. Install them one by one if needed.

## Running the Project

### Training

```bash
conda activate founddiff
CUDA_VISIBLE_DEVICES=1 python train.py --name FoundDiff --is_train --train_num_steps 400000
```

### Inference

Download the model weights from the Google Drive link above, then run:

```bash
conda activate founddiff
CUDA_VISIBLE_DEVICES=4 python train.py --name FoundDiff --epoch 400 --dataset 2020_seen
```
Please refer to options files for more setting.

Ensure:
- `DA-CLIP.pth` is placed in `src/` (or as specified in DA-Diff.py)
- `model-400.pt` is placed in `checkpoints/FoundDiff/sample/`

## Environment Summary

| Package | Version | Source |
|---------|---------|--------|
| Python | 3.10 | conda |
| PyTorch | 2.1.0+cu118 | pytorch index |
| torchvision | 0.16.0+cu118 | pytorch index |
| numpy | 1.26.x | pip |
| CUDA Toolkit | 11.8.0 | nvidia/label/cuda-11.8.0 |
| causal-conv1d | 1.2.2.post1 | pip (pre-built) |
| mamba-ssm | 2.2.2 | GitHub releases |

## Data Format and Structure

### Required Data Format

The code expects data in **NumPy .npy format** (not PNG, JPG, or DICOM). Each file should be a 2D numpy array of shape (512, 512).

```python
# How data is loaded in the code (from pdf_dataset.py)
q_data = np.load(self.q_path_list[index]).astype(np.float32) # Low dose CT
f_data = np.load(self.f_path_list[index]).astype(np.float32) # Full dose CT
```

### Required Folder Structure

The code expects data organized in this structure:

#### For Mayo 2020 Dataset (Testing)

```
Mayo2020_ab_2d/test/full_1mm/ # High dose (full dose) - abdominal
Mayo2020_ab_2d/test/quarter_1mm/ # Low dose (1/4 dose) - abdominal

Mayo2020_lung_2d/test/full_1mm/ # High dose - lung
Mayo2020_lung_2d/test/quarter_1mm/ # Low dose - lung

Mayo2020_head_2d_2/test/full_1mm/ # High dose - head
Mayo2020_head_2d_2/test/quarter_1mm/ # Low dose - head
```

#### For Mayo 2016 Dataset

```
Mayo2016_2d/test/full_1mm/ # High dose
Mayo2016_2d/test/quarter_1mm/ # Low dose
```

Or alternatively:
```
Mayo2016_2d/train/full_1mm/ # High dose (training)
Mayo2016_2d/quarter_1mm/ # Low dose
```

### Files to Modify for Custom Data Paths

You need to modify paths in **2 files**:

#### 1. data/pdf_dataset.py (Lines 331-399)

Key paths to change:
```python
# Line 331-340: Abdominal data
ab_ndct = sorted_list('/mnt/miah203/zhchen/Mayo2020_ab_2d/'+ phase+'/full_1mm/*')[:num]
ab_dose_1_4_list = sorted_list('/mnt/miah203/zhchen/Mayo2020_ab_2d/'+phase+'/quarter_1mm/*')[start:num:stride]

# Line 354-363: Lung data
lung_ndct = sorted_list('/mnt/miah203/zhchen/Mayo2020_lung_2d/'+ phase+'/full_1mm/*')[:num]
lung_dose_1_4_list = sorted_list('/mnt/miah203/zhchen/Mayo2020_lung_2d/'+phase+'/quarter_1mm/*')[start:num:stride]

# Line 382-391: Head data
head_ndct = sorted_list('/mnt/miah203/zhchen/Mayo2020_head_2d_2/'+ phase+'/full_1mm/*')[:num]
head_dose_1_4_list = sorted_list('/mnt/miah203/zhchen/Mayo2020_head_2d_2/'+phase+'/quarter_1mm/*')[start:num:stride]

# Line 398-399: Mayo2016 data
self.mayo16_ldct = sorted_list('/mnt/miah203/zhchen/Mayo2016_2d/quarter_1mm/*')
self.mayo16_ndct_path_list = sorted_list('/mnt/miah203/zhchen/Mayo2016_2d/train/full_1mm/*')
```

#### 2. data/mayo16_dataset.py (Lines 42-49)

```python
# Line 42-43: Test data paths
self.f_path_list = sorted_list('/mnt/miah203/zhchen/CQ500_2d/test/full_1mm/*')
self.q_path_list = sorted_list('/mnt/miah203/zhchen/CQ500_2d/test/sim-0.25/*')

# Line 46-48: Training data paths
self.q_path_list = sorted_list('/mnt/miah203/zhchen/Mayo2016_2d/test/quarter_1mm/*')
self.f_path_list = sorted_list('/mnt/miah203/zhchen/Mayo2016_2d/test/full_1mm/*')
```

### Example: Setting Custom Data Path

To use your own data at `/home/user/my_data/`, change the paths in pdf_dataset.py:

```python
# Before:
ab_ndct = sorted_list('/mnt/miah203/zhchen/Mayo2020_ab_2d/'+ phase+'/full_1mm/*')[:num]

# After:
ab_ndct = sorted_list('/home/user/my_data/Mayo2020_ab_2d/'+ phase+'/full_1mm/*')[:num]
```

### Data Filename Matching

The code matches low-dose and high-dose files by index:
- Low dose file: `ab-001.npy` → loads from `quarter_1mm/`
- High dose file: `001.npy` → loads from `full_1mm/` using index from filename

Ensure your low-dose and high-dose files have matching indices in their filenames.

### Converting DICOM to .npy

If your data is in DICOM format (.IMA, .dcm), you need to convert it:

```python
import numpy as np
import pydicom
from PIL import Image
import os

def dicom_to_npy(dicom_path, output_path):
"""Convert DICOM to .npy format"""
ds = pydicom.dcmread(dicom_path)
img = ds.pixel_array

# Resize to 512x512 if needed
if img.shape != (512, 512):
img = Image.fromarray(img).resize((512, 512))
img = np.array(img)

# Normalize to HU values if needed
# img = img * ds.RescaleSlope + ds.RescaleIntercept

np.save(output_path, img.astype(np.float32))

# Example usage:
for f in os.listdir('/path/to/dicom/files'):
if f.endswith('.dcm') or f.endswith('.IMA'):
dicom_to_npy(f, f.replace('.dcm', '.npy'))
```

## Troubleshooting

### "libc10.so: cannot open shared object file"
- Ensure the activation script in Step 6 is created correctly
- Or manually run: `export LD_LIBRARY_PATH="$CONDA_PREFIX/lib/python3.10/site-packages/torch/lib:$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"`

### "nvcc shows CUDA 13.2"
- You installed cuda-toolkit without specifying the channel label
- Recreate environment and use: `conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit -y`

### mamba-ssm compilation fails
- Always use pre-built wheels from GitHub releases
- Building from source requires specific toolchain and is not recommended

## Credits

- FoundDiff: "Foundational Diffusion Model for Generalizable Low-Dose CT Denoising"
- Mamba SSM: https://github.com/state-spaces/mamba
Loading