app.py

import streamlit as st
from PIL import Image
import jax
import numpy as np
import jax.numpy as jnp  # JAX NumPy
from flax.training import train_state  # Useful dataclass to keep train state
from flax import linen as nn  # Linen API
from huggingface_hub import HfFileSystem
from flax.serialization import msgpack_restore, from_state_dict
import os

class CNN(nn.Module):
  """A simple CNN model."""

  @nn.compact
  def __call__(self, x):
    x = nn.Conv(features=32, kernel_size=(3, 3))(x)
    x = nn.relu(x)
    x = nn.avg_pool(x, window_shape=(2, 2), strides=(2, 2))
    x = nn.Conv(features=64, kernel_size=(3, 3))(x)
    x = nn.relu(x)
    x = nn.avg_pool(x, window_shape=(2, 2), strides=(2, 2))
    x = x.reshape((x.shape[0], -1))  # flatten
    x = nn.Dense(features=256)(x)
    x = nn.relu(x)
    x = nn.Dense(features=10)(x)
    return x

cnn = CNN()
params = cnn.init(jax.random.PRNGKey(0), jnp.ones([1, 28, 28, 1]))['params']

fs = HfFileSystem()
with fs.open("PrakhAI/HelloWorld/checkpoint.msgpack", "rb") as f:
    params = from_state_dict(params, msgpack_restore(f.read())["params"])

uploaded_files = st.file_uploader("Input Images of handwritten digit (example in files)", type=['jpg','png','tif'], accept_multiple_files=True)

if len(uploaded_files) == 0:
    st.write("Please upload an image!")
else:
  input = jnp.array([jnp.array(Image.open(uploaded_file).convert("HSV").split()[2].resize((28, 28))).reshape(1, 28, 28, 1) / 255. for uploaded_file in uploaded_files])
  predictions = cnn.apply({"params": params}, input)
  for (image, prediction) in zip(uploaded_files, predictions):
    st.image(Image.open(image))
    st.write(f"Model Prediction: {prediction.argmax().tolist()}")


def gridify(kernel, grid, kernel_size, scaling=5, padding=1):
    scaled_and_padded = np.pad(np.repeat(np.repeat(kernel, repeats=scaling, axis=0), repeats=scaling, axis=1), ((padding,),(padding,),(0,),(0,)), 'constant', constant_values=(-1,))
    grid = np.pad(np.array(scaled_and_padded.reshape((kernel_size[0]*scaling+2*padding, kernel_size[1]*scaling+2*padding, grid[0], grid[1])).transpose(2,0,3,1).reshape(grid[0]*(kernel_size[0]*scaling+2*padding), grid[1]*(kernel_size[1]*scaling+2*padding))+1)*127., (padding,), 'constant', constant_values=(0,))
    st.image(Image.fromarray(np.repeat(np.expand_dims(grid, axis=0), repeats=3, axis=0).astype(np.uint8).transpose(1,2,0), mode="RGB"))

with st.expander("See first convolutional layer"):
    gridify(params["Conv_0"]["kernel"], grid=(4,8), kernel_size=(3,3))
    
with st.expander("See second convolutional layer"):
    print(params["Conv_1"]["kernel"].shape)
    gridify(params["Conv_1"]["kernel"], grid=(32,64), kernel_size=(3,3))