Update README.md (#25)
Browse files- Update README.md (1f0515555b87f739d1ad02dcc7580e764c540171)
README.md
CHANGED
|
@@ -46,7 +46,190 @@ BiomedCLIP establishes new state of the art in a wide range of standard datasets
|
|
| 46 |
|
| 47 |
## Model Use
|
| 48 |
|
| 49 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
Please refer to this [example notebook](https://aka.ms/biomedclip-example-notebook).
|
| 52 |
|
|
|
|
| 46 |
|
| 47 |
## Model Use
|
| 48 |
|
| 49 |
+
### 1. Environment
|
| 50 |
+
|
| 51 |
+
```bash
|
| 52 |
+
conda create -n biomedclip python=3.10 -y
|
| 53 |
+
conda activate biomedclip
|
| 54 |
+
pip install open_clip_torch==2.23.0 transformers==4.35.2 matplotlib
|
| 55 |
+
```
|
| 56 |
+
|
| 57 |
+
### 2.1 Load from HF hub
|
| 58 |
+
|
| 59 |
+
```python
|
| 60 |
+
import torch
|
| 61 |
+
from urllib.request import urlopen
|
| 62 |
+
from PIL import Image
|
| 63 |
+
from open_clip import create_model_from_pretrained, get_tokenizer
|
| 64 |
+
|
| 65 |
+
# Load the model and config files from the Hugging Face Hub
|
| 66 |
+
model, preprocess = create_model_from_pretrained('hf-hub:microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224')
|
| 67 |
+
tokenizer = get_tokenizer('hf-hub:microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224')
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
# Zero-shot image classification
|
| 71 |
+
template = 'this is a photo of '
|
| 72 |
+
labels = [
|
| 73 |
+
'adenocarcinoma histopathology',
|
| 74 |
+
'brain MRI',
|
| 75 |
+
'covid line chart',
|
| 76 |
+
'squamous cell carcinoma histopathology',
|
| 77 |
+
'immunohistochemistry histopathology',
|
| 78 |
+
'bone X-ray',
|
| 79 |
+
'chest X-ray',
|
| 80 |
+
'pie chart',
|
| 81 |
+
'hematoxylin and eosin histopathology'
|
| 82 |
+
]
|
| 83 |
+
|
| 84 |
+
dataset_url = 'https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224/resolve/main/example_data/biomed_image_classification_example_data/'
|
| 85 |
+
test_imgs = [
|
| 86 |
+
'squamous_cell_carcinoma_histopathology.jpeg',
|
| 87 |
+
'H_and_E_histopathology.jpg',
|
| 88 |
+
'bone_X-ray.jpg',
|
| 89 |
+
'adenocarcinoma_histopathology.jpg',
|
| 90 |
+
'covid_line_chart.png',
|
| 91 |
+
'IHC_histopathology.jpg',
|
| 92 |
+
'chest_X-ray.jpg',
|
| 93 |
+
'brain_MRI.jpg',
|
| 94 |
+
'pie_chart.png'
|
| 95 |
+
]
|
| 96 |
+
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
|
| 97 |
+
model.to(device)
|
| 98 |
+
model.eval()
|
| 99 |
+
|
| 100 |
+
context_length = 256
|
| 101 |
+
|
| 102 |
+
images = torch.stack([preprocess(Image.open(urlopen(dataset_url + img))) for img in test_imgs]).to(device)
|
| 103 |
+
texts = tokenizer([template + l for l in labels], context_length=context_length).to(device)
|
| 104 |
+
with torch.no_grad():
|
| 105 |
+
image_features, text_features, logit_scale = model(images, texts)
|
| 106 |
+
|
| 107 |
+
logits = (logit_scale * image_features @ text_features.t()).detach().softmax(dim=-1)
|
| 108 |
+
sorted_indices = torch.argsort(logits, dim=-1, descending=True)
|
| 109 |
+
|
| 110 |
+
logits = logits.cpu().numpy()
|
| 111 |
+
sorted_indices = sorted_indices.cpu().numpy()
|
| 112 |
+
|
| 113 |
+
top_k = -1
|
| 114 |
+
|
| 115 |
+
for i, img in enumerate(test_imgs):
|
| 116 |
+
pred = labels[sorted_indices[i][0]]
|
| 117 |
+
|
| 118 |
+
top_k = len(labels) if top_k == -1 else top_k
|
| 119 |
+
print(img.split('/')[-1] + ':')
|
| 120 |
+
for j in range(top_k):
|
| 121 |
+
jth_index = sorted_indices[i][j]
|
| 122 |
+
print(f'{labels[jth_index]}: {logits[i][jth_index]}')
|
| 123 |
+
print('\n')
|
| 124 |
+
```
|
| 125 |
+
|
| 126 |
+
### 2.2 Load from local files
|
| 127 |
+
|
| 128 |
+
```python
|
| 129 |
+
import json
|
| 130 |
+
|
| 131 |
+
from urllib.request import urlopen
|
| 132 |
+
from PIL import Image
|
| 133 |
+
import torch
|
| 134 |
+
from huggingface_hub import hf_hub_download
|
| 135 |
+
from open_clip import create_model_and_transforms, get_tokenizer
|
| 136 |
+
from open_clip.factory import HF_HUB_PREFIX, _MODEL_CONFIGS
|
| 137 |
+
|
| 138 |
+
|
| 139 |
+
# Download the model and config files
|
| 140 |
+
hf_hub_download(
|
| 141 |
+
repo_id="microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224",
|
| 142 |
+
filename="open_clip_pytorch_model.bin",
|
| 143 |
+
local_dir="checkpoints"
|
| 144 |
+
)
|
| 145 |
+
hf_hub_download(
|
| 146 |
+
repo_id="microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224",
|
| 147 |
+
filename="open_clip_config.json",
|
| 148 |
+
local_dir="checkpoints"
|
| 149 |
+
)
|
| 150 |
+
|
| 151 |
+
|
| 152 |
+
# Load the model and config files
|
| 153 |
+
model_name = "biomedclip_local"
|
| 154 |
+
|
| 155 |
+
with open("checkpoints/open_clip_config.json", "r") as f:
|
| 156 |
+
config = json.load(f)
|
| 157 |
+
model_cfg = config["model_cfg"]
|
| 158 |
+
preprocess_cfg = config["preprocess_cfg"]
|
| 159 |
+
|
| 160 |
+
|
| 161 |
+
if (not model_name.startswith(HF_HUB_PREFIX)
|
| 162 |
+
and model_name not in _MODEL_CONFIGS
|
| 163 |
+
and config is not None):
|
| 164 |
+
_MODEL_CONFIGS[model_name] = model_cfg
|
| 165 |
+
|
| 166 |
+
tokenizer = get_tokenizer(model_name)
|
| 167 |
+
|
| 168 |
+
model, _, preprocess = create_model_and_transforms(
|
| 169 |
+
model_name=model_name,
|
| 170 |
+
pretrained="checkpoints/open_clip_pytorch_model.bin",
|
| 171 |
+
**{f"image_{k}": v for k, v in preprocess_cfg.items()},
|
| 172 |
+
)
|
| 173 |
+
|
| 174 |
+
|
| 175 |
+
# Zero-shot image classification
|
| 176 |
+
template = 'this is a photo of '
|
| 177 |
+
labels = [
|
| 178 |
+
'adenocarcinoma histopathology',
|
| 179 |
+
'brain MRI',
|
| 180 |
+
'covid line chart',
|
| 181 |
+
'squamous cell carcinoma histopathology',
|
| 182 |
+
'immunohistochemistry histopathology',
|
| 183 |
+
'bone X-ray',
|
| 184 |
+
'chest X-ray',
|
| 185 |
+
'pie chart',
|
| 186 |
+
'hematoxylin and eosin histopathology'
|
| 187 |
+
]
|
| 188 |
+
|
| 189 |
+
dataset_url = 'https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224/resolve/main/example_data/biomed_image_classification_example_data/'
|
| 190 |
+
test_imgs = [
|
| 191 |
+
'squamous_cell_carcinoma_histopathology.jpeg',
|
| 192 |
+
'H_and_E_histopathology.jpg',
|
| 193 |
+
'bone_X-ray.jpg',
|
| 194 |
+
'adenocarcinoma_histopathology.jpg',
|
| 195 |
+
'covid_line_chart.png',
|
| 196 |
+
'IHC_histopathology.jpg',
|
| 197 |
+
'chest_X-ray.jpg',
|
| 198 |
+
'brain_MRI.jpg',
|
| 199 |
+
'pie_chart.png'
|
| 200 |
+
]
|
| 201 |
+
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
|
| 202 |
+
model.to(device)
|
| 203 |
+
model.eval()
|
| 204 |
+
|
| 205 |
+
context_length = 256
|
| 206 |
+
|
| 207 |
+
images = torch.stack([preprocess(Image.open(urlopen(dataset_url + img))) for img in test_imgs]).to(device)
|
| 208 |
+
texts = tokenizer([template + l for l in labels], context_length=context_length).to(device)
|
| 209 |
+
with torch.no_grad():
|
| 210 |
+
image_features, text_features, logit_scale = model(images, texts)
|
| 211 |
+
|
| 212 |
+
logits = (logit_scale * image_features @ text_features.t()).detach().softmax(dim=-1)
|
| 213 |
+
sorted_indices = torch.argsort(logits, dim=-1, descending=True)
|
| 214 |
+
|
| 215 |
+
logits = logits.cpu().numpy()
|
| 216 |
+
sorted_indices = sorted_indices.cpu().numpy()
|
| 217 |
+
|
| 218 |
+
top_k = -1
|
| 219 |
+
|
| 220 |
+
for i, img in enumerate(test_imgs):
|
| 221 |
+
pred = labels[sorted_indices[i][0]]
|
| 222 |
+
|
| 223 |
+
top_k = len(labels) if top_k == -1 else top_k
|
| 224 |
+
print(img.split('/')[-1] + ':')
|
| 225 |
+
for j in range(top_k):
|
| 226 |
+
jth_index = sorted_indices[i][j]
|
| 227 |
+
print(f'{labels[jth_index]}: {logits[i][jth_index]}')
|
| 228 |
+
print('\n')
|
| 229 |
+
|
| 230 |
+
```
|
| 231 |
+
|
| 232 |
+
### Use in Jupyter Notebook
|
| 233 |
|
| 234 |
Please refer to this [example notebook](https://aka.ms/biomedclip-example-notebook).
|
| 235 |
|