Initial Commit

.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
+pretrain_vit_feature_extractor_food101.pth filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,69 @@
+
+### Imports and class names setup ---------------------------------------------------- ###
+import os
+import torch
+import torchvision
+import gradio as gr
+
+from model import create_vit
+from timeit import default_timer as timer
+from typing import Tuple, Dict
+
+# Setup class names
+with open("class_names.txt", "r") as f:
+  class_names = [food.strip() for food in f.readlines()]
+
+# Device agnostic code
+if torch.backends.mps.is_available():
+  device = 'mps'
+elif torch.cuda.is_available():
+  device = 'cuda'
+else:
+  device = 'cpu'
+
+### Model and transforms preparation ---------------------------------------------------- ###
+vit_model, vit_transforms = create_vit(pretrained_weights=torchvision.models.ViT_B_16_Weights.DEFAULT,
+                                       model=torchvision.models.vit_b_16,
+                                       in_features=768,
+                                       out_features=101,
+                                       device='cpu')
+
+# Load save weights
+vit_model.load_state_dict(torch.load(f="pretrained_vit_feature_extractor_food101.pth",
+                          map_location=torch.device("cpu"))) # load the model to the CPU
+
+### Predict function ---------------------------------------------------- ###
+def predict(img) -> Tuple[Dict, float]:
+  # Start a timer
+  start_time = timer()
+  # Transform the input image for use with ViT Model
+  img = vit_transforms(img).unsqueeze(0) # unsqueeze = add batch dimension on 0th index (3, 224, 224) into (1, 3, 224, 224)
+  # Put model into eval mode, make prediction
+  vit_model.eval()
+  with torch.inference_mode():
+    # Pass transformed image through the model and turn the prediction logits into probabilities
+    pred_logits = vit_model(img)
+    pred_probs = torch.softmax(pred_logits, dim=1)
+  # Create a prediction label and prediction probability dictionary
+  pred_labels_and_probs = {class_names[i]: float(pred_probs[0][i]) for i in range(len(class_names))}
+  
+  # Calculate pred time
+  end_timer = timer()
+  pred_time = round(end_timer - start_time, 4)
+  
+  # Return pred dict and pred time
+  return pred_labels_and_probs, pred_time
+
+### Gradio interface and launch ------------------------------------------------------------------ ###
+
+# Create title and description
+title = "FoodVision: ViT Model"
+description = "A ViT model trained on 20% of the Food101 dataset to classify Food images"
+
+# Create example list
+example_list = [["examples/" + example] for example in os.listdir("examples")]
+
+# Create the Gradio demo
+demo = gr.Interface(fn=predict, inputs=gr.Image(type="pil"), outputs=[gr.Label(num_top_classes=5, label="Predictions"),
+                                                         gr.Number(label="Prediction time(s)")], title=title, description=description, examples=example_list)
+demo.launch()

class_names.txt

@@ -0,0 +1,101 @@
+apple_pie
+baby_back_ribs
+baklava
+beef_carpaccio
+beef_tartare
+beet_salad
+beignets
+bibimbap
+bread_pudding
+breakfast_burrito
+bruschetta
+caesar_salad
+cannoli
+caprese_salad
+carrot_cake
+ceviche
+cheese_plate
+cheesecake
+chicken_curry
+chicken_quesadilla
+chicken_wings
+chocolate_cake
+chocolate_mousse
+churros
+clam_chowder
+club_sandwich
+crab_cakes
+creme_brulee
+croque_madame
+cup_cakes
+deviled_eggs
+donuts
+dumplings
+edamame
+eggs_benedict
+escargots
+falafel
+filet_mignon
+fish_and_chips
+foie_gras
+french_fries
+french_onion_soup
+french_toast
+fried_calamari
+fried_rice
+frozen_yogurt
+garlic_bread
+gnocchi
+greek_salad
+grilled_cheese_sandwich
+grilled_salmon
+guacamole
+gyoza
+hamburger
+hot_and_sour_soup
+hot_dog
+huevos_rancheros
+hummus
+ice_cream
+lasagna
+lobster_bisque
+lobster_roll_sandwich
+macaroni_and_cheese
+macarons
+miso_soup
+mussels
+nachos
+omelette
+onion_rings
+oysters
+pad_thai
+paella
+pancakes
+panna_cotta
+peking_duck
+pho
+pizza
+pork_chop
+poutine
+prime_rib
+pulled_pork_sandwich
+ramen
+ravioli
+red_velvet_cake
+risotto
+samosa
+sashimi
+scallops
+seaweed_salad
+shrimp_and_grits
+spaghetti_bolognese
+spaghetti_carbonara
+spring_rolls
+steak
+strawberry_shortcake
+sushi
+tacos
+takoyaki
+tiramisu
+tuna_tartare
+waffles

model.py

@@ -0,0 +1,28 @@
+
+import torch
+import torchvision
+
+from torch import nn
+
+def create_vit(pretrained_weights: torchvision.models.Weights,
+               model: torchvision.models,
+               in_features: int,
+               out_features: int,
+               device: torch.device):
+    """Creates a Vision Transformer (ViT) instance from torchvision
+    and returns it.
+    """
+    # Create a pretrained ViT model
+    model = torchvision.models.vit_b_16(weights=pretrained_weights).to(device)
+    transforms = pretrained_weights.transforms()
+
+    # Freeze the feature extractor
+    for param in model.parameters():
+        param.requires_grad = False
+
+    # Change the head of the ViT
+    model.heads = nn.Sequential(
+        nn.Linear(in_features=in_features, out_features=out_features)
+    ).to(device)
+
+    return model, transforms

pretrained_vit_feature_extractor_food101.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8539aefc12da9f3198ae140021e430d49c0091bbd244a7621b9534a79e8d19ae
+size 343568235

requirements.txt

@@ -0,0 +1,4 @@
+
+torch==2.0.1
+torchvision==0.15.2
+gradio==3.23.0