Upload 4 files

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 +1,72 @@
+"""
+Ryan Tietjen
+Aug 2024
+Creates a vit base 16 model for the demo
+"""
+
+import torch
+import torchvision
+from torch import nn
+
+def vit_b_16(num_classes:int=101, 
+             seed:int=31, 
+             freeze_gradients:bool=True,
+             unfreeze_blocks=0):
+    """
+    Initializes and configures a Vision Transformer (ViT-B/16) model with options for freezing gradients
+    and adjusting the number of trainable blocks.
+
+    This function sets up a ViT-B/16 model pre-trained on the ImageNet-1K dataset, modifies the classification
+    head to accommodate a specified number of classes, and optionally freezes the gradients of certain blocks
+    to prevent them from being updated during training.
+
+    Parameters:
+    num_classes (int): The number of output classes for the new classification head. Default is 101.
+    seed (int): Random seed for reproducibility. Default is 31.
+    freeze_gradients (bool): If True, freezes the gradients of the model's parameters, except for the last few
+                             blocks specified by `unfreeze_blocks`. Default is True.
+    unfreeze_blocks (int): Number of transformer blocks from the end whose parameters will have trainable gradients.
+                           Default is 0, implying all are frozen except the new classification head.
+
+    Returns:
+    tuple: A tuple containing:
+        - model (torch.nn.Module): The modified ViT-B/16 model with a new classifier head.
+        - transforms (callable): The transformation function required for input images, as recommended by the
+                                 pre-trained weights.
+
+    Example:
+    ```python
+    model, transform = vit_b_16(num_classes=101, seed=31, freeze_gradients=True, unfreeze_blocks=2)
+    ```
+
+    Notes:
+    - The total number of parameters in the model is calculated and used to determine which parameters to freeze.
+    - The classifier head of the model is replaced with a new linear layer that outputs to the specified number of classes.
+    """
+
+    torch.manual_seed(seed) 
+
+    #Create model and extract weights/transforms
+    weights = torchvision.models.ViT_B_16_Weights.IMAGENET1K_SWAG_E2E_V1
+    transforms = weights.transforms()
+    model = torchvision.models.vit_b_16(weights=weights)
+
+    params = list(model.parameters())
+    params_to_unfreeze = 4 + (12 * unfreeze_blocks)
+    # Total number of parameters
+    total_params = len(params)
+
+
+    #Freeze gradients to avoid modifying the original model    
+    if freeze_gradients:
+        for i, param in enumerate(params):
+            # Set requires_grad to False for all but the last n encoder blocks
+            if i < total_params - params_to_unfreeze:
+                param.requires_grad = False
+
+    #modify classifier model to fit our 
+    model.heads = nn.Sequential(
+        nn.Linear(in_features=768,
+                  out_features=num_classes))
+    
+    return model, transforms

requirements.txt

@@ -0,0 +1,4 @@
+torch==1.12.0
+torchvision==0.13.0
+gradio==4.40.0
+numpy==1.26.4

vit_b_16_unfreeze_one_encoder_block_10_total_epochs.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a91f23cd7a2b19958a8fb041786f083180d0dbac3242867ae900f4e0db185d45
+size 344740050