tensorflow + pygame 手写数字识别的小游戏

起因，目的:

很久之前，一个客户的作业，我帮忙写的。
今天删项目，觉得比较简洁，发出来给大家看看。

效果图:

在这里插入图片描述

1. 训练模型的代码

import sys
import tensorflow as tf# Use MNIST handwriting dataset
mnist = tf.keras.datasets.mnist# Prepare data for training
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
y_train = tf.keras.utils.to_categorical(y_train)
y_test = tf.keras.utils.to_categorical(y_test)
x_train = x_train.reshape(x_train.shape[0], x_train.shape[1], x_train.shape[2], 1
)
x_test = x_test.reshape(x_test.shape[0], x_test.shape[1], x_test.shape[2], 1
)"""
model = tf.keras.models.Sequential([tf.keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),tf.keras.layers.MaxPooling2D((2, 2)),tf.keras.layers.Flatten(),tf.keras.layers.Dense(128, activation='relu'),tf.keras.layers.Dropout(0.2),tf.keras.layers.Dense(10, activation='softmax')
])"""# Create a convolutional neural network
model = tf.keras.models.Sequential([# 1.  Convolutional layer. Learn 32 filters using a 3x3 kernel, activation function is relu, input shape (28,28,1)tf.keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),#2. Max-pooling layer, using 2x2 pool sizetf.keras.layers.MaxPooling2D((2, 2)),#3.  Flatten unitstf.keras.layers.Flatten(),#4. Add a hidden layer with dropout,tf.keras.layers.Dropout(0.2),#5. Add an output layer with output units for all 10 digits, activation function is softmaxtf.keras.layers.Dense(10, activation='softmax')])# Train neural network
model.compile(optimizer="adam",loss="categorical_crossentropy",metrics=["accuracy"]
)
model.fit(x_train, y_train, epochs=10)# Evaluate neural network performance
model.evaluate(x_test,  y_test, verbose=2)# Save model to file
if len(sys.argv) == 2:filename = sys.argv[1]model.save(filename)print(f"Model saved to {filename}.")"""
Run this code:  python handwriting.py model_1.pthoutput:1875/1875 [==============================] - 10s 5ms/step - loss: 0.0413 - accuracy: 0.9873
Epoch 8/10
1875/1875 [==============================] - 10s 5ms/step - loss: 0.0385 - accuracy: 0.9877
Epoch 9/10
1875/1875 [==============================] - 10s 5ms/step - loss: 0.0338 - accuracy: 0.9898
Epoch 10/10
1875/1875 [==============================] - 10s 5ms/step - loss: 0.0319 - accuracy: 0.9900
313/313 - 1s - loss: 0.0511 - accuracy: 0.9845 - 718ms/epoch - 2ms/step
Model saved to model_1.pth.
"""

2. 运行小游戏，进行识别

从命令行运行:

python recognition.py model.h5


import numpy as np
import pygame
import sys
import tensorflow as tf
import time"""
run this code:
python recognition.py  model_1.pth or,  python recognition.py  model.h5 output:
"""print("len(sys.argv): ", len(sys.argv))# Check command-line arguments
if len(sys.argv) != 2:print("Usage: python recognition.py model")sys.exit()model = tf.keras.models.load_model(sys.argv[1])# Colors
BLACK = (0, 0, 0)
WHITE = (255, 255, 255)# Start pygame
pygame.init()
size = width, height = 600, 400
screen = pygame.display.set_mode(size)# Fonts
OPEN_SANS = "assets/fonts/OpenSans-Regular.ttf"
smallFont = pygame.font.Font(OPEN_SANS, 20)
largeFont = pygame.font.Font(OPEN_SANS, 40)ROWS, COLS = 28, 28OFFSET = 20
CELL_SIZE = 10handwriting = [[0] * COLS for _ in range(ROWS)]
classification = Nonewhile True:# Check if game quitfor event in pygame.event.get():if event.type == pygame.QUIT:sys.exit()screen.fill(BLACK)# Check for mouse pressclick, _, _ = pygame.mouse.get_pressed()if click == 1:mouse = pygame.mouse.get_pos()else:mouse = None# Draw each grid cellcells = []for i in range(ROWS):row = []for j in range(COLS):rect = pygame.Rect(OFFSET + j * CELL_SIZE,OFFSET + i * CELL_SIZE,CELL_SIZE, CELL_SIZE)# If cell has been written on, darken cellif handwriting[i][j]:channel = 255 - (handwriting[i][j] * 255)pygame.draw.rect(screen, (channel, channel, channel), rect)# Draw blank cellelse:pygame.draw.rect(screen, WHITE, rect)pygame.draw.rect(screen, BLACK, rect, 1)# If writing on this cell, fill in current cell and neighborsif mouse and rect.collidepoint(mouse):handwriting[i][j] = 250 / 255if i + 1 < ROWS:handwriting[i + 1][j] = 220 / 255if j + 1 < COLS:handwriting[i][j + 1] = 220 / 255if i + 1 < ROWS and j + 1 < COLS:handwriting[i + 1][j + 1] = 190 / 255# Reset buttonresetButton = pygame.Rect(30, OFFSET + ROWS * CELL_SIZE + 30,100, 30)resetText = smallFont.render("Reset", True, BLACK)resetTextRect = resetText.get_rect()resetTextRect.center = resetButton.centerpygame.draw.rect(screen, WHITE, resetButton)screen.blit(resetText, resetTextRect)# Classify buttonclassifyButton = pygame.Rect(150, OFFSET + ROWS * CELL_SIZE + 30,100, 30)classifyText = smallFont.render("Classify", True, BLACK)classifyTextRect = classifyText.get_rect()classifyTextRect.center = classifyButton.centerpygame.draw.rect(screen, WHITE, classifyButton)screen.blit(classifyText, classifyTextRect)# Reset drawingif mouse and resetButton.collidepoint(mouse):handwriting = [[0] * COLS for _ in range(ROWS)]classification = None# Generate classificationif mouse and classifyButton.collidepoint(mouse):classification = model.predict([np.array(handwriting).reshape(1, 28, 28, 1)]).argmax()# Show classification if one existsif classification is not None:classificationText = largeFont.render(str(classification), True, WHITE)classificationRect = classificationText.get_rect()grid_size = OFFSET * 2 + CELL_SIZE * COLSclassificationRect.center = (grid_size + ((width - grid_size) / 2),100)screen.blit(classificationText, classificationRect)pygame.display.flip()