Update README

HieuPhan33 · Mar 31, 2024 · 5a7d821 · 5a7d821
2 parents 2ac4bc5 + a9ededf
commit 5a7d821
Show file tree

Hide file tree

Showing 7 changed files with 33 additions and 17 deletions.
diff --git a/README.md b/README.md
@@ -16,8 +16,8 @@ If you find our work useful, please cite our paper.
 
 ```
 @article{phan2024decomposing,
-    title={Decomposing Disease Descriptions for Enhanced Pathology Detection: A Multi-Aspect  Vision-Language Matching Framework}, 
-    author={Minh Hieu Phan and Yutong Xie and Yuankai Qi and Lingqiao Liu and Liyang Liu and Bowen Zhang and Zhibin Liao and Qi Wu and Minh-Son To and Johan W. Verjans},
+    title={Decomposing Disease Descriptions for Enhanced Pathology Detection: A Multi-Aspect  Vision-Language Pre-training Framework}, 
+    author={Vu Minh Hieu Phan and Yutong Xie and Yuankai Qi and Lingqiao Liu and Liyang Liu and Bowen Zhang and Zhibin Liao and Qi Wu and Minh-Son To and Johan W. Verjans},
     year={2024},
     journal={arXiv preprint arXiv:2403.07636},
 }
@@ -225,9 +225,9 @@ Our pre-train code is given in ```Pretrain```.
 
 * Run `accelerate launch --multi_gpu --num_processes=4 --num_machines=1 --num_cpu_threads_per_process=8 train_MAVL.py --root /data/2019.MIMIC-CXR-JPG/2.0.0 --config configs/MAVL_resnet.yaml --bs 124 --num_workers 8`
 
-Note: The reported results in our paper are obtained by pre-training on 4 x A100 for 60 epochs. We provided the checkpoint [here](Pretrain/data_file/DATA_Prepare.md).
+Note: The reported results in our paper are obtained by pre-training on 4 x A100 for 60 epochs. We provided the checkpoints [here](Pretrain/data_file/DATA_Prepare.md). We found that ckpts at later stage (`checkpoint_full_46.pth`) yields higher zero-shot classification accuracy. Ckpt at earlier stage (`checkpoint_full_40.pth`) yields more stable accuracy on visual grounding.
 
-We also conducted a lighter pre-training schedule with 2 x A100 for 40 epochs using mixed precision training, achieving similar zero-shot classification results. Checkpoints for this setup are also available [here](Pretrain/data_file/DATA_Prepare.md).
+We also conducted a lighter pre-training schedule with 2 x A100 for 40 epochs using mixed precision training, achieving similar zero-shot classification results. Checkpoint for this setup is also available [here](Pretrain/data_file/DATA_Prepare.md).
 
 ```
 accelerate launch --multi_gpu --num_processes=2 --num_machines=1 --num_cpu_threads_per_process=8 --mixed_precision=fp16 train_MAVL.py --root /data/2019.MIMIC-CXR-JPG/2.0.0 --config configs/MAVL_short.yaml --bs 124 --num_workers 8

diff --git a/Zero-shot_classification/configs/chexpert_mavl.yaml b/Zero-shot_classification/configs/chexpert_mavl.yaml
@@ -1,11 +1,14 @@
-test_file: '/data/data/chexpertv1/CheXpert-v1.0/filter_test_labels.csv'
+test_file: '/data/VLM/data/chexpertv1/CheXpert-v1.0/filter_test_labels.csv'
 disease_book: 'observation explanation.json'
 concept_book: '../Pretrain/concept_gen/gpt4_mimic.json'
 
 dataset: 'chexpert'
-root: '/data/data/chexpertv1'
+root: '/data/VLM/data/chexpertv1'
 model: 'mavl'
-model_path: '../checkpoints/checkpoint_full_40.pth'
+model_path: '../checkpoints/checkpoint_full_46.pth'
+# model_path: '../../medslip/checkpoints/cl_global_1e-1_true/checkpoint_40.pth'
+
+
 mode: feature
 
 image_res: 224

diff --git a/Zero-shot_classification/configs/padchest_seen_mavl.yaml b/Zero-shot_classification/configs/padchest_seen_mavl.yaml
@@ -1,16 +1,17 @@
-test_file: '/data/data/PadChest/test.csv'
+test_file: '/data/VLM/data/PadChest/test.csv'
 disease_book: 'observation explanation_padchest_rare.json'
 # concept_book: '../Pretrain/concept_gen/gpt4_mimic_padchest_rare.json'
 concept_book: '../Pretrain/concept_gen/gpt4_mimic_padchest_rare.json'
 
 dataset: 'padchest'
-root: '/data/data/PadChest/'
+root: '/data/VLM/data/PadChest/'
 model: 'mavl'
-model_path: '../checkpoints/checkpoint_short_37.pth'
+# model_path: '../checkpoints/checkpoint_short_37.pth'
+model_path: '../../medslip/checkpoints/cl_global_1e-1_true/checkpoint_40.pth'
 
 
 image_res: 224
-test_batch_size: 128
+test_batch_size: 512
 
 d_model: 256
 base_model: 'resnet50'

diff --git a/Zero-shot_classification/configs/rsna_mavl.yaml b/Zero-shot_classification/configs/rsna_mavl.yaml
@@ -1,12 +1,11 @@
-test_file: '/data/data/rsna/test.csv'
+test_file: '/data/VLM/data/rsna/test.csv'
 disease_book: 'observation explanation.json'
 concept_book: '../Pretrain/concept_gen/gpt4_mimic.json'
 
 dataset: 'rsna'
-root: '/data/data/rsna'
+root: '/data/VLM/data/rsna'
 model: mavl
-# model_path: '../checkpoints/checkpoint_full_40.pth'
-model_path: '../../medslip/checkpoints/cl_global_1e-1_true/checkpoint_40.pth'
+model_path: '../checkpoints/checkpoint_full_46.pth'
 
 
 image_res: 224

diff --git a/Zero-shot_classification/result_mavl_chexpert_feature.csv b/Zero-shot_classification/result_mavl_chexpert_feature.csv
@@ -0,0 +1,7 @@
+Class Name,Accuracy,Max F1,AUC ROC,Precision,Recall
+cardiomegaly,0.818,0.7513812154696132,0.9042486574697812,0.6445497630331753,0.9006622516556292
+edema,0.886,0.6666666666666666,0.9210110584518167,0.6222222222222222,0.717948717948718
+consolidation,0.932,0.4,0.8639724723625449,0.4230769230769231,0.3793103448275862
+atelectasis,0.796,0.7247956403269755,0.8752330903542973,0.6214953271028038,0.869281045751634
+effusion,0.902,0.7798165137614679,0.9506847319347319,0.7456140350877193,0.8173076923076923
+Average,0.8668000000000001,0.6645320072449447,0.9030300021146346,0.6113916541045687,0.736902010498252
diff --git a/Zero-shot_classification/result_mavl_rsna_feature.csv b/Zero-shot_classification/result_mavl_rsna_feature.csv
@@ -0,0 +1,3 @@
+Class Name,Accuracy,Max F1,AUC ROC,Precision,Recall
+pneumonia,0.8081319093123478,0.6237587348289814,0.863606256560271,0.5649566955363091,0.6962233169129721
+Average,0.8081319093123478,0.6237587348289814,0.863606256560271,0.5649566955363091,0.6962233169129721
diff --git a/Zero-shot_grounding/configs/covid_mavl.yaml b/Zero-shot_grounding/configs/covid_mavl.yaml
@@ -1,11 +1,14 @@
-test_file: '/data/data/covid_rural/test.csv'
+test_file: '/data/VLM/data/covid_rural/test.csv'
 disease_book: 'observation explanation.json'
 concept_book: '../Pretrain/concept_gen/gpt4_mimic_covidr.json'
+# concept_book: '../../medslip/Pretrain/concept_gen/gpt4_mimic_covid.json'
 
 dataset: 'covid-r'
-root: '/data/data/covid_rural'
+root: '/data/VLM/data/covid_rural'
 target: 'covid19'
 model_path: '../checkpoints/checkpoint_full_40.pth'
+# model_path: '../../medslip/checkpoints/cl_global_1e-1_true/checkpoint_40.pth'
+
 
 image_res: 224
 test_batch_size: 512