ridhimagarg / diffusion

ChestXNet

This repository is based on openai/guided-diffusion.

Download pre-trained models

We have released checkpoints for the main models in the paper. Before using these models, please review the corresponding model card to understand the intended use and limitations of these models.

Here are the download links for each model trained from scratch checkpoint:

• Model 1 diffusion model-1.pt
• Model 2 diffusion model-2.pt
• Model 3 diffusion model-3.pt
• Model 4 diffusion model-4.pt
• Model 5 diffusion model-5.pt
• Classifier classifier.pt

Here are the download links for finetuned model

• Model 1 diffusion model-1.pt

Same classifier model is used for finetuning

Sampling from pre-trained models

To sample from these models, you can use the classifier_sample.py, image_sample.py scripts. Here, we provide flags for sampling from all of these models. We assume that you have downloaded the relevant model checkpoints into a folder called models/.

For these examples, we will generate 100 samples with batch size 4. Feel free to change these values.

SAMPLE_FLAGS="--batch_size 4 --num_samples 100 --timestep_respacing 250"

Classifier guidance

Note for these sampling runs that you can set --classifier_scale 0 to sample from the base diffusion model. You may also use the image_sample.py script instead of classifier_sample.py in that case.

• Model 1 model:

MODEL_FLAGS="--image_size 256 --num_channels 128 --num_res_blocks 3 --class_cond True --num_heads 4 --attention_resolutions 16,8" 
DIFFUSION_FLAGS="--diffusion_steps 4000 --noise_schedule linear"
CLASSIFIER_FLAGS="--image_size 256 --classifier_attention_resolutions 32,16,8 --classifier_depth 2 --classifier_width 128 --classifier_pool attention --classifier_resblock_updown True --classifier_use_scale_shift_norm True"

python scripts/classifier_sample.py --model_path {model1_path}/model090000.pt --classifier_path {classifier_path}/model085000.pt $MODEL_FLAGS $DIFFUSION_FLAGS $SAMPLE_FLAGS $CLASSIFIER_FLAGS

Evaluation of diffusion model

It is presented in this [readme] (guided-diffusion/evaluations/README.md)

Classification abormality Pleural Effusion

For this module, first we have to convert the synthetic data into the original(CheXpert) data format using this [script] (guided-diffusion/scripts/convert_generated_data_chexpert_format.py)

The code for classification is in classifier folder.

To use the pretrained classifier and test on the new synthetic data the following command needs to run -:

python classifier_predict.py {dataset name} densenet121-res224-all

eg -: python classifier_predict.py chex densenet121-res224-all

To finetune model, we used the train.csv generated from this [script] (guided-diffusion/scripts/convert_generated_data_chexpert_format.py) instead of test.csv

Results

This table summarizes our model trained from scratch results for pleural effusion pathology:

Dataset	FID	sFID	FID (XRV)
Model-1	65.95	79.69	4.87
Model-2	446.36	154.64	10.92
Model-3	440.38	309.43	10.03
Model-4	105.01	90.80	5.89
Model-5	96.24	105.41	5.27
Model(18k)	30.13	75.94	1.35

This table summarizes our model trained for multiple pathologies:

Dataset	FID	sFID
Atelectasis (Infected)	109.33	350.12
Atelectasis (Healthy)	118.73	365.16
Cardiomegaly (Infected)	105.03	107.95
Cardiomegaly (Healthy)	107.95	362.55

Training models

Training diffusion models is described in the parent repository.