add final code

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+convnext_model.pth filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,6 +1,6 @@
 ---
 title: DogVision
-emoji: 🏢
+emoji: 🐶
 colorFrom: gray
 colorTo: indigo
 sdk: gradio

app.py

@@ -0,0 +1,91 @@
+### 1. Imports and class names setup ###
+from model import create_model
+import pandas as pd
+import torch
+from typing import Tuple, Dict
+from timeit import default_timer as timer
+import gradio as gr
+import os
+import numpy as np
+
+### 1. Create a Dictionary for Dog Breeds
+labels_csv = pd.read_csv('./labels.csv')
+labels = labels_csv['breed']
+labels = np.array(labels)
+unique_labels = np.unique(labels)
+
+### 2. Model and transforms preparation ###
+model, model_transforms = create_model(num_classes=len(unique_labels))
+model = torch.compile(model)
+
+
+# Load save weights
+model.load_state_dict(torch.load(f='./convnext_model.pth', map_location='cpu',weights_only=True))
+
+# 3. Predict Function
+
+def predict(img) -> Tuple[Dict[str, float], str]:
+    """
+    Predicts the class probabilities for a given image using a pre-trained model.
+
+    Args:
+        img: A PIL image to be predicted.
+
+    Returns:
+        A tuple containing:
+        - A formatted string displaying class labels and their respective probabilities.
+        - The time taken for inference in seconds as a string.
+    """
+    # Start a timer
+    start_time = timer()
+
+    # Put the model into evaluation mode and disable gradient computation
+    model.eval()
+    with torch.inference_mode():
+        
+        # Transform the input image for use with the model
+        img = model_transforms(img).unsqueeze(dim=0)
+        
+        # Pass transformed image through the model
+        pred_logit = model(img)
+
+    # Turn prediction logits into probabilities
+    pred_prob = torch.softmax(pred_logit, dim=1)
+    
+    pred_label = torch.argmax(pred_prob, dim=1)
+
+    # Map probabilities to class labels
+    prediction = unique_labels[pred_label]
+    probabilities = {unique_labels[i]: pred_prob[0, i].item() for i in range(len(unique_labels))}
+
+    # Calculate the time taken
+    end_time = timer()
+    inference_time = end_time - start_time
+
+    # Return predictions as a dictionary and inference time
+    return probabilities, f"{inference_time:.4f} seconds"
+### 4. Gradio app ###
+
+
+# Create title, description and article
+title = "Dogvision 🐶"
+description = "A [ConvNeXt Tiny](https://pytorch.org/vision/stable/models/generated/torchvision.models.convnext_tiny.html#torchvision.models.convnext_tiny) Computer Vision Model To Classify 120 Dog Breeds 🐩 Ranging fro A Labrador 🐕 to A German Shepherd! 🐕‍🦺"
+article = "Created with 🤎 (and a mixture of mathematics, statistics, and tons of calculations 👩🏽‍🔬) by Arpit Vaghela [GitHub](https://github.com/magnifiques)"
+
+# Create example list
+example_list = [["./examples/" + example] for example in os.listdir("examples")]
+
+demo = gr.Interface(fn=predict,
+                    inputs=gr.Image(type='pil'),
+                    outputs=[
+                    gr.Label(num_top_classes=3, label="Top Predictions"),  # Display top predictions with probabilities
+                    gr.Textbox(label="Prediction Time (s)")  # Display inference time
+                    ],
+                    examples=example_list,
+                    title=title,
+                    description=description,
+                    article=article)
+
+# Launch the demo!
+demo.launch(debug=False, # print errors locally?
+            share=True) # generate a publicly shareable URL

breeds.txt

@@ -0,0 +1,120 @@
+boston_bull
+dingo
+pekinese
+bluetick
+golden_retriever
+bedlington_terrier
+borzoi
+basenji
+scottish_deerhound
+shetland_sheepdog
+walker_hound
+maltese_dog
+norfolk_terrier
+african_hunting_dog
+wire-haired_fox_terrier
+redbone
+lakeland_terrier
+boxer
+doberman
+otterhound
+standard_schnauzer
+irish_water_spaniel
+black-and-tan_coonhound
+cairn
+affenpinscher
+labrador_retriever
+ibizan_hound
+english_setter
+weimaraner
+giant_schnauzer
+groenendael
+dhole
+toy_poodle
+border_terrier
+tibetan_terrier
+norwegian_elkhound
+shih-tzu
+irish_terrier
+kuvasz
+german_shepherd
+greater_swiss_mountain_dog
+basset
+australian_terrier
+schipperke
+rhodesian_ridgeback
+irish_setter
+appenzeller
+bloodhound
+samoyed
+miniature_schnauzer
+brittany_spaniel
+kelpie
+papillon
+border_collie
+entlebucher
+collie
+malamute
+welsh_springer_spaniel
+chihuahua
+saluki
+pug
+malinois
+komondor
+airedale
+leonberg
+mexican_hairless
+bull_mastiff
+bernese_mountain_dog
+american_staffordshire_terrier
+lhasa
+cardigan
+italian_greyhound
+clumber
+scotch_terrier
+afghan_hound
+old_english_sheepdog
+saint_bernard
+miniature_pinscher
+eskimo_dog
+irish_wolfhound
+brabancon_griffon
+toy_terrier
+chow
+flat-coated_retriever
+norwich_terrier
+soft-coated_wheaten_terrier
+staffordshire_bullterrier
+english_foxhound
+gordon_setter
+siberian_husky
+newfoundland
+briard
+chesapeake_bay_retriever
+dandie_dinmont
+great_pyrenees
+beagle
+vizsla
+west_highland_white_terrier
+kerry_blue_terrier
+whippet
+sealyham_terrier
+standard_poodle
+keeshond
+japanese_spaniel
+miniature_poodle
+pomeranian
+curly-coated_retriever
+yorkshire_terrier
+pembroke
+great_dane
+blenheim_spaniel
+silky_terrier
+sussex_spaniel
+german_short-haired_pointer
+french_bulldog
+bouvier_des_flandres
+tibetan_mastiff
+english_springer
+cocker_spaniel
+rottweiler

convnext_model.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f99258fdd9b2f3a2e137dff009a5b1bc3aaea5e2b349f644fd3d03e5ec8c27c
+size 111721116

model.py

@@ -0,0 +1,55 @@
+import torch
+import torch.nn as nn
+import torchvision
+from torchvision import transforms
+
+# Custom transformation to handle palette images
+def convert_to_rgba(image):
+    # Check if the image mode is 'P' (palette mode)
+    if image.mode == 'P':
+        image = image.convert('RGBA')
+    return image
+
+def create_model(num_classes: int = 120, seed: int = 42):
+
+
+    # 1. Download the default weights
+    weights = torchvision.models.ConvNeXt_Tiny_Weights.IMAGENET1K_V1
+
+    # 2. Setup transforms
+    default_transforms = weights.transforms()
+
+    custom_transforms = transforms.Compose([
+    # transforms.RandomHorizontalFlip(p=0.5),      # Randomly flip images horizontally
+    # transforms.Lambda(convert_to_rgba),          # Apply RGBA conversion if necessary
+    # transforms.RandomRotation(degrees=10),       # Randomly rotate images by up to 10 degrees
+    # transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1),  # Color jitter
+    ])
+
+    # 3. Combine custom and ViT's default transforms
+    combined_transforms = transforms.Compose([
+      custom_transforms,     # First, apply your custom augmentations
+      transforms.Resize((224, 224)),  # Resize to ConvNext's input size if needed (ConvNext expects 224x224)
+      transforms.ToTensor(),  # Convert image to Tensor
+      default_transforms,         # Apply default normalization (mean, std)
+     ])
+
+    # 4. Create a model and apply the default weights
+    model = torchvision.models.convnext_tiny(weights=weights)
+
+    # 5. Freeze the base layers in the model (this will stop all layers from training)
+    for parameters in model.parameters():
+        parameters.requires_grad = False
+
+    # 6. Set seeds for reproducibility
+    torch.manual_seed(seed)
+
+    # 7. Modify the number of output layers (add a dropout layer for regularization)
+    model.classifier = nn.Sequential(
+      nn.LayerNorm([768, 1, 1], eps=1e-06, elementwise_affine=True),  # Apply LayerNorm on the channel dimension (768)
+      nn.Flatten(start_dim=1),  # Flatten the tensor from dimension 1 onwards (batch size remains intact)
+      nn.Linear(in_features=768, out_features=num_classes, bias=True)
+    )
+
+
+    return model, combined_transforms

requirements.txt

@@ -0,0 +1,4 @@
+torch==2.4.0
+torchvision==0.19.0
+gradio==4.44.0
+triton==3.1.0