| | import gradio as gr |
| | import torch |
| | import open_clip |
| | import joblib |
| |
|
| | def predict(input_image): |
| | if input_image is None: |
| | return "No image submitted..." |
| | |
| | device = torch.device("cpu") |
| | |
| | |
| | |
| | model, _, preprocess = open_clip.create_model_and_transforms('ViT-L-14', pretrained='openai', device=device) |
| |
|
| | image = preprocess(input_image).unsqueeze(0).to(device) |
| |
|
| | with torch.amp.autocast(device_type=device.type): |
| | with torch.no_grad(): |
| | image_features = model.encode_image(image) |
| | image_features /= image_features.norm(dim=-1, keepdim=True) |
| |
|
| | embedding = image_features[0].cpu().float().numpy() |
| |
|
| | model = joblib.load('model.pkl') |
| | result = model.predict([embedding]) |
| |
|
| | return "Map" if result == 1 else "No map" |
| |
|
| | demo = gr.Interface(fn=predict, |
| | inputs=gr.Image(label="Input image", type="pil"), |
| | outputs="text", |
| | title="MapPool model", |
| | description="The model predicts whether an image is a map or not. It takes about 30 seconds since it runs on a CPU (it is much faster on a GPU). Although the validation accuracy of the model is 98.5%, some outputs may not be correct. In this case, feel free to contact <a href='https://schnuerer.dev/contact'>me</a>.", |
| | article="More information: <a href='https://huggingface.co/datasets/sraimund/MapPool'>MapPool - Bubbling up an extremely large corpus of maps for AI</a><br>Keywords: map identification, map recognition, map classification") |
| | demo.launch() |
