Error while training for rock paper scissors

TensorFlow General Discussion
,
1

So I’m trying my hands on Convolution Neural Network and thought a more hands-on approach is the way to go with the rock paper scissors dataset

This is how the model has been built

model = tf.keras.models.Sequential([
    tf.keras.layers.Conv2D(filters=32,
                          kernel_size=3,
                          activation="relu",
                          input_shape=(200, 200, 3)),
    tf.keras.layers.Conv2D(32, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(pool_size=2, padding="valid"),
    tf.keras.layers.Dropout(0.2),
    
    tf.keras.layers.Conv2D(64, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(pool_size=2, padding="valid"),
    tf.keras.layers.Dropout(0.2),
    
    tf.keras.layers.Conv2D(128, 3, activation="relu"),
    tf.keras.layers.MaxPooling2D(pool_size=2, padding="valid"),
    tf.keras.layers.Dropout(0.2),
    
    tf.keras.layers.Flatten(),
    
    tf.keras.layers.Dense(256, activation="relu"),
    tf.keras.layers.Dropout(0.5),
    
    tf.keras.layers.Dense(3, activation="softmax")
])

And this is the summary

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_6 (Conv2D)            (None, 198, 198, 32)      896       
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 196, 196, 32)      9248      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 98, 98, 32)        0         
_________________________________________________________________
dropout_4 (Dropout)          (None, 98, 98, 32)        0         
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 96, 96, 64)        18496     
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 48, 48, 64)        0         
_________________________________________________________________
dropout_5 (Dropout)          (None, 48, 48, 64)        0         
_________________________________________________________________
conv2d_9 (Conv2D)            (None, 46, 46, 128)       73856     
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 23, 23, 128)       0         
_________________________________________________________________
dropout_6 (Dropout)          (None, 23, 23, 128)       0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 67712)             0         
_________________________________________________________________
dense_2 (Dense)              (None, 256)               17334528  
_________________________________________________________________
dropout_7 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 3)                 771       
=================================================================
Total params: 17,437,795
Trainable params: 17,437,795
Non-trainable params: 0
_________________________________________________________________

The model has been compiled as such

model.compile(loss="categorical_crossentropy",
             optimizer=tf.optimizers.Adam(),
             metrics=['accuracy'])

and trained

history = model.fit(train_data, epochs=10, steps_per_epoch=20, validation_data = val_data, validation_steps=5, verbose=1)

But while training I get thrown this error

InvalidArgumentError: Input to reshape is a tensor with 3686400 values, but the requested shape requires a multiple of 67712
[[node sequential_1/flatten_1/Reshape (defined at tmp/ipykernel_17/154995288.py:1) ]] [Op:__inference_train_function_2119]
Function call stack:
train_function

What am I doing wrong? Any help is greatly appreciated

3

The link says that the data has shape (200,300) but your input shape says (200,200). I don’t know if you resized your input to the latter but you may consider changing it to the appropriate size before training. It may solve your issue.

5

changing the input shape of the first Conv2D’s input_shape=(300,200,3), resulted in the same error

7

You mean (200,300,3)?

9

Tensorflow takes data in the format (H,W,C) = (200,300,3)