input images are stored locally

mnist_classifier.ipynb

@@ -536,7 +536,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 59,
+   "execution_count": 60,
    "metadata": {
     "tags": [
      "exclude"
@@ -547,14 +547,96 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "[NbConvertApp] Converting notebook mnist_classifier.ipynb to script\n"
+      "[NbConvertApp] Converting notebook mnist_classifier.ipynb to script\n",
+      "[NbConvertApp] Writing 3568 bytes to mnist_classifier.py\n"
      ]
     }
    ],
    "source": [
     "!jupyter nbconvert --to script --TagRemovePreprocessor.remove_cell_tags=\"exclude\" --TemplateExporter.exclude_input_prompt=True mnist_classifier.ipynb\n"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": 78,
+   "metadata": {
+    "tags": [
+     "exclude"
+    ]
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "tensor(0.1046) tensor(1.) tensor(0.) tensor(0.3062)\n",
+      "tensor(0.1435) tensor(1.) tensor(0.) tensor(0.3220)\n"
+     ]
+    },
+    {
+     "data": {
+      "image/png": "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",
+      "text/plain": [
+       "<Figure size 144x144 with 1 Axes>"
+      ]
+     },
+     "metadata": {
+      "needs_background": "light"
+     },
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "from PIL import Image\n",
+    "import numpy as np\n",
+    "import glob\n",
+    "import os\n",
+    "png_files = glob.glob(os.path.join('./uploads', '*png'))\n",
+    "img = png_files[1]\n",
+    "image = Image.open(img)\n",
+    "image = TF.to_tensor(image).permute(1, 2, 0)\n",
+    "print(image.mean(), image.max(), image.min(), image.std())\n",
+    "print(xb[0].mean(), xb[0].max(), xb[0].min(), xb[0].std())\n",
+    "plt.imshow(image);\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 75,
+   "metadata": {
+    "tags": [
+     "exclude"
+    ]
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "<matplotlib.image.AxesImage at 0x7ff734b20fd0>"
+      ]
+     },
+     "execution_count": 75,
+     "metadata": {},
+     "output_type": "execute_result"
+    },
+    {
+     "data": {
+      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAI4AAACOCAYAAADn/TAIAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAAAI70lEQVR4nO3df2yU9R0H8PenpaURcR2MKhRSmrQguoUxGwpO0f1gqYyMzc1Ig8RtZGwB58RlQZmzJiabmWZsjE3jImEowS2gwzmMkc7JfhiBTTYQrDAIcqWzdmOuwlZ77Wd/9Gjv89Ber5/79dz1/UrI3ee55+l9Y95+73vPPfc5UVUQjVRRrgdA+YnBIRcGh1wYHHJhcMiFwSGXlIIjIg0i0iIix0TkrnQNisJPvOdxRKQYwBsAFgKIANgHoFFVD6dveBRWY1I4di6AY6p6HABE5EkASwAMGZxSGatlGJfCU1K2deJMh6pOCm5PJTiVAE7F1REA9YkOKMM41MsnUnhKyrbduv3kYNtTCY4Msu2C1z0RWQlgJQCU4aIUno7CJJXFcQTAtLh6KoDTwZ1U9VFVrVPVuhKMTeHpKExSCc4+ALUiUi0ipQCWAngmPcOisHO/VKlqVERuA/A8gGIAm1T1tbSNjEItlTUOVHUXgF1pGgvlEZ45JhcGh1wYHHJhcMiFwSEXBodcGBxyYXDIhcEhFwaHXBgccmFwyIXBIRcGh1wYHHJhcMglpQu5aHAdX51v6s6PnTX1znmPmHpWqb2If853V5m6YuOf0ji69OCMQy4MDrkwOOTCNU6Szt1ov6Ta8aHi/vvLP99sHvtS+YOmrii2a5jewPfLurXH1Gu/vs3UW5o/aQfT+pb9e+fOmVqjUWQaZxxyYXDIhcEhF65xYsZMrTR1ZOMlpt5T90NTXywD65TeQK+F585VmLpp/RcTPvfZBe+aekPdL+xYbrBdRn67ZrOp5+25zdQ1tx7sv5+p9Q5nHHJhcMiFwSEXdw9Aj0tkgoalI5dePdvU7933jqmbr3wq4fGvd3f137/xiTvNYzWPvGnqaKQ14d8qKisz9fjdtt3dtuoXEh7/z97/mnrZLQNrnqKXXk147HB26/Y/q2pdcDtnHHIZNjgisklE2kXkUNy2CSLygogcjd2+P7PDpLBJZsbZDKAhsO0uAM2qWgugOVbTKDLseRxV3SMi0wOblwC4Pnb/5wB+B2BtOgeWbsE1zdrHnzD1grL3AkfY3phz/7LU1Jd9beDcy/TWl81jw505kbH2s6rWJ6tN/Wr1VlMHzxO19dg1zU33fMvU5S/Z8WSCd41zqaq2AUDstmKY/anAZPzMMdvVFibvjPOWiEwGgNht+1A7sl1tYfLOOM8AuBXAA7HbnWkbUYa019lzIxeuaaw1bfb6m4qbI6aOBq6BiVc8cYKpOxbPNPVDTQ+bev7Y4DXFdn21t8vWX95q1zTTt2R+TROUzNvxbQBeBjBTRCIisgJ9gVkoIkfR9yMgD2R2mBQ2ybyrahzioXCcAqac4JljcinY63HGVE0z9YY7fjqi4/e2V5m6e/kFv7zT71/13aa+s95+trSqfLepg+dlgtp77PqpacXtpp7+YvbXNEGccciFwSEXBodcCnaNoxfZa1zGSfC8TTES+f1se90v7EddKIo71zLcmmXw34Qb2nWB8zTVIVjTBHHGIRcGh1wK9qWq58hRUzedXGLqnbW/SenvP955Wf/9//WWmMe+8r5Twd2Nd3u7TH3V02tMPfP+v5q61zPADOOMQy4MDrkwOORSsGucoO6P20uGPl08L+H+ZxqvMvXEX9nfqe25fOAjiXfus5dyrphtvx4TidrHm04vMnXt7a+YOoxrmiDOOOTC4JALg0Muo2aNg17bLk0DdVB54HJMqZxi6mVbnuu/3zjetlYL+k7rYlO/ffW/E+6fDzjjkAuDQy4MDrmMnjVOij747GlTLxsff17IXjZx8/FPmbrz2o5MDStnOOOQC4NDLgwOuXCNc16RvZT0+PfmmvrpST8KHDDwn65LbWOTll/PMPUUcI1DBIDBIScGh1y4xonpaviIqQ/fsjGwh10DxX8l5spnbUv8Gd8P308hphtnHHJJpj/ONBF5UUSOiMhrIvKN2Ha2rB3FkplxogC+qaqzAMwDsFpErgBb1o5qyTRWagNwvsNop4gcAVCJPGxZG+/sF2yrtp89tN7URbBfIQ6qv3d1//1ZT71uHkt8pU9hGNEaJ9bveA6AV8CWtaNa0sERkYsB7ABwh6r+ZwTHrRSR/SKyvxtdwx9AeSGp4IhICfpCs1VVz/+sSlIta9mutjANu8YREQHwGIAjqvqDuIfyqmVt7zUfNvX8dXtNXVNiQx1sXbIqssDUl+460X8/euZMGkaYX5I5AfhRAMsBHBSRA7Ft69AXmF/G2te+CeCmjIyQQimZd1V/wNCdgdiydpTimWNyKdjPqnqut5893fPYZlNfWxb8cSA7qV6+dbWpa+4/ZOrezn+kNL58xxmHXBgccmFwyKVg1jhSUmrqGQ8eNvWFaxrrj132/6EZ60+YOtrZmcLoCg9nHHJhcMilYF6qOj87x9Q/nvLwEHv22XHWXne2YZ39ld9xbba9GlmccciFwSEXBodcCmaNU77ffgTwuWO2JeyOGtuC/94DnzF11XauaUaCMw65MDjkwuCQS8GscaInTtr6Ovv4YtgW+1U4mOkhFTTOOOTC4JALg0MuDA65MDjkwuCQC4NDLqKqw++VricTeRvASQAfAELbw5Vjs6pUdVJwY1aD0/+kIvtVtS7rT5wEji05fKkiFwaHXHIVnEdz9LzJ4NiSkJM1DuU/vlSRS1aDIyINItIiIsdEJKftbUVkk4i0i8ihuG2h6N2cD72lsxYcESkG8BMANwC4AkBjrF9yrmwG0BDYFpbezeHvLa2qWfkHYD6A5+PquwHcna3nH2JM0wEciqtbAEyO3Z8MoCWX44sb104AC8M0vmy+VFUCOBVXR2LbwiR0vZvD2ls6m8EZrI8g39Il4O0tnQ3ZDE4EwLS4eiqA00PsmytJ9W7OhlR6S2dDNoOzD0CtiFSLSCmApejrlRwm53s3Azns3ZxEb2kg172ls7zIWwTgDQB/B/DtHC84t6Hvx0260TcbrgAwEX3vVo7GbifkaGzXoO9l/G8ADsT+LQrL+FSVZ47Jh2eOyYXBIRcGh1wYHHJhcMiFwSEXBodcGBxy+T+hXTkLJRIRbwAAAABJRU5ErkJggg==",
+      "text/plain": [
+       "<Figure size 144x144 with 1 Axes>"
+      ]
+     },
+     "metadata": {
+      "needs_background": "light"
+     },
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "plt.imshow(xb[0].permute(1,2,0))"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,

mnist_classifier.py

@@ -48,7 +48,7 @@ def conv(ni, nf, ks=3, s=2, act=nn.ReLU, norm=None):
 
 def _conv_block(ni, nf, ks=3, s=2, act=nn.ReLU, norm=None):
     return nn.Sequential(
-        conv(ni, nf, ks=ks, s=1, norm=norm, act=act),
+        conv(ni, nf, ks=ks, s=1, norm=None, act=act),
         conv(nf, nf, ks=ks, s=s, norm=norm, act=act),
     )
 
@@ -77,15 +77,26 @@ def forward(self, x):
 
 def cnn_classifier():
     return nn.Sequential(
-        ResBlock(1, 8,),
-        ResBlock(8, 16, ),
-        ResBlock(16, 32,),
-        ResBlock(32, 64, ),
-        ResBlock(64, 64,),
+        ResBlock(1, 8, norm=nn.LayerNorm([8, 14, 14])),
+        ResBlock(8, 16, norm=nn.LayerNorm([16, 7, 7])),
+        ResBlock(16, 32, norm=nn.LayerNorm([32, 4, 4])),
+        ResBlock(32, 64, norm=nn.LayerNorm([64, 2, 2])),
+        ResBlock(64, 64,  norm=nn.LayerNorm([64, 1, 1])),
         conv(64, 10, act=False),
         nn.Flatten(),
     )
 
+# def cnn_classifier():
+#     return nn.Sequential(
+#         ResBlock(1, 8,),
+#         ResBlock(8, 16, ),
+#         ResBlock(16, 32,),
+#         ResBlock(32, 64, ),
+#         ResBlock(64, 64,),
+#         conv(64, 10, act=False),
+#         nn.Flatten(),
+#     )
+
 
 def kaiming_init(m):
     if isinstance(m, (nn.Conv1d, nn.Conv2d, nn.Conv3d)):

server.py

@@ -1,12 +1,13 @@
-from fastapi import FastAPI, File, UploadFile
-import io
+from fastapi import FastAPI, UploadFile
 from PIL import Image
 from fastapi.responses import FileResponse
 from fastapi.staticfiles import StaticFiles
 from pathlib import Path
-import torchvision.transforms as transforms
 import mnist_classifier
 import torch
+from pathlib import Path
+import datetime
+import numpy as np
 
 app = FastAPI()
 
@@ -15,20 +16,29 @@
 async def root():
     return FileResponse("static/index.html")
 
-def process_image(file: UploadFile):
-    image_bytes = file.file.read()
-    pil_image = Image.open(io.BytesIO(image_bytes))
-    transform = transforms.Compose([
-        transforms.Resize((28, 28)),
-        transforms.Grayscale(num_output_channels=1),
-        transforms.ToTensor(),
-    ])
-    tensor_image = transform(pil_image)
-    return tensor_image
+upload_dir = Path("uploads")
+upload_dir.mkdir(parents=True, exist_ok=True)
 
+def process_image(file):
+    image = Image.open(file.file)
+    image = image.resize((28, 28))  # Resize to MNIST image size
+    image = image.convert("L")  # Convert to grayscale
+    raw_image = image
+    image = np.array(image)
+    image = image / 255.0  # Normalize pixel values
+    return torch.from_numpy(image).float().reshape(1, 28, 28), raw_image
+
+def store_img(image):
+    timestamp = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
+    unique_filename = f"{timestamp}.png"
+    output_path = upload_dir / unique_filename
+    image.save(output_path)
+
 @app.post("/predict")
 async def predict(image: UploadFile):
-    tensor_image = process_image(image)
+    tensor_image, raw_image = process_image(image)
+    print(tensor_image.shape)
     prediction = mnist_classifier.predict(tensor_image)
+    store_img(raw_image)
     return {"prediction": prediction}