how to obtain input.json

Hey, I’m new to zkML. I want to try compiling my Keras model into Circom, and I found this repository. Great work!

I’ve successfully obtained circuit.circom, circuit.json, circuit.wasm, and generate_witness.js. However, I’m now struggling with creating the witness because I don’t understand what input.json should look like.

My model is LeNet-5, with the following configuration:

def build_lenet5_relu(input_shape=(32, 32, 1), num_classes=4):
    model = models.Sequential()

    # Input layer
    model.add(layers.Input(shape=input_shape))

    # First Convolutional Layer
    model.add(layers.Conv2D(6, kernel_size=(5, 5), strides=(1, 1), padding='same', activation='relu'))
    model.add(layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2)))

    # Second Convolutional Layer
    model.add(layers.Conv2D(16, kernel_size=(5, 5), strides=(1, 1), activation='relu'))
    model.add(layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2)))

    # Third Convolutional Layer
    model.add(layers.Conv2D(120, kernel_size=(5, 5), strides=(1, 1), activation='relu'))

    # Flattening Layer
    model.add(layers.Flatten())

    # Second Fully Connected Layer
    model.add(layers.Dense(num_classes, activation='softmax'))

    model.output_names=['output']

    return model

I assumed that I should merge your circom.json with "in" as the input, so I did:

image = Image.open(<input-image-path>)
image_array = np.array(image)

x = np.expand_dims(image_array, axis=0)

data_array = x.reshape([32, 32, 1]).tolist()

with open(<path-to-circuit-json>, 'r') as circuit_file:
    input_json = json.load(circuit_file)

input_json["in"] = data_array

with open(<path-to-input-json>, 'w') as input_file:
    json.dump(input_json, input_file, indent=4)

But when I generate the witness using generate_witness.js, I get this error:

Error: Not all inputs have been set. Only 53640 out of 85137

I suspect there might be an issue with the signals in my input.json, probably wrong format?. Do you have any clue?

Thanks for your response.

ora-io / keras2circom

how to obtain input.json #7