import gradio as gr from fastai.vision.all import * def is_cat(x): return x[0].isupper() def get_labels(path): pet_type = 'cat' if is_cat(path.name) else 'dog' breed = RegexLabeller(r'(.+)_\d+.jpg$')(path.name) return [pet_type, breed] learn = load_learner('dog_cat_multi_v3.pkl') def predict(image): # Transform the image pil_image = PILImage.create(image) # Predict preds, mask, probs = learn.predict(pil_image) # Apply the threshold threshold = 0.570 classes_with_probs = [(learn.dls.vocab[i], probs[i].item()) for i in range(len(mask)) if probs[i] > threshold] confidences = {learn.dls.vocab[i]: float(probs[i].item()) for i in range(len(probs)) if probs[i] > threshold} # Check if the prediction includes "dog" or "cat" pet_type = None breed = 'unknown' # Default to 'unknown' if no breed is identified pet_prob = 0 breed_prob = 0 for class_name, prob in classes_with_probs: if class_name == 'dog' or class_name == 'cat': pet_type = class_name pet_prob = prob else: breed = class_name breed_prob = prob # Check if pet_type is None (i.e., neither dog nor cat) if pet_type is None: return "This is not a cat, nor a dog." result = f"This is a {pet_type}, the breed is {breed if breed else 'unknown'}.\n" \ f"The probability for it being a {pet_type} is {pet_prob * 100:.2f}%, the probability of being the breed is {breed_prob * 100:.2f}%." return result, confidences # Define the Gradio interface iface = gr.Interface( fn=predict, inputs=gr.inputs.Image(shape=(224, 224)), outputs=[gr.outputs.Textbox(label="Prediction"), gr.outputs.Label(label='confidences',num_top_classes=2)], theme=gr.themes.Soft(), examples=["images/cazou103_Generate_a_high-resolution_image_of_a_Beagle_showcasin_609a1bae-22ac-4158-9091-dbf3220b2765.PNG", "images/cazou103_Generate_a_high-resolution_image_of_a_Shiba_Inu_showca_998f6162-289f-450e-ab02-558c3a575f61.PNG", "images/cazou103_Generate_a_high-resolution_image_of_a_Siamese_cat_show_68b1d78f-c304-4164-9342-4a4b56d2c4c5.PNG" ], title="Cat and Dog Image Classifier", description="Upload an image of a cat or a dog, and the model will identify the type and breed.", article="This model has been trained on the Oxford Pets dataset and might not recognize all types dog and cat breeds. For best results, use clear images of pets." ) # Launch the interface iface.launch()