DevHeads IoT Integration Server•3mo ago

How do I fix a tensor dimension mismatch in TinyML disease detection?

Hello guys am workin on Disease Detection from X-Ray Scans Using TinyML, i have gathered a diverse dataset of X-ray images from public medical databases, used images labeled with specific diseases or conditions, such as pneumonia, tuberculosis, or normal/healthy cases, i have also prepared my training script but keep getting an error while training the model

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "tensorflow/lite/python/interpreter.py", line 42, in set_tensor
    self._interpreter.SetTensor(self._tensor_index_map[tensor_index], value)
ValueError: Cannot set tensor: Dimension mismatch. Got [1, 128, 128, 3], expected [1, 64, 64, 1]

here's my code

import tensorflow as tf
from tensorflow.keras import layers, models

model = models.Sequential([
    layers.Conv2D(16, (3, 3), activation='relu', input_shape=(64, 64, 1)),
    layers.MaxPooling2D((2, 2), name="pool_1"),
    layers.Conv2D(32, (3, 3), activation='relu', name="conv_2"),
    layers.MaxPooling2D((2, 2), name="pool_2"),
    layers.Flatten(name="flatten"),
    layers.Dense(64, activation='relu', name="dense_1"),
    layers.Dense(3, activation='softmax', name="output_layer")
])

import tensorflow as tf
from tensorflow.keras import layers, models

model = models.Sequential([
    layers.Conv2D(16, (3, 3), activation='relu', input_shape=(64, 64, 1)),
    layers.MaxPooling2D((2, 2), name="pool_1"),
    layers.Conv2D(32, (3, 3), activation='relu', name="conv_2"),
    layers.MaxPooling2D((2, 2), name="pool_2"),
    layers.Flatten(name="flatten"),
    layers.Dense(64, activation='relu', name="dense_1"),
    layers.Dense(3, activation='softmax', name="output_layer")
])

2 Replies

RED HAT•3mo ago

Hello @Enthernet Code The error you’re encountering is due to a mismatch between the input image dimensions expected by your model and the actual dimensions of the images being fed into it. Your model expects images with a shape of (64, 64, 1) (grayscale), but the images you’re providing have a shape of (128, 128, 3) (colored). To resolve this, you have two options: 1. Preprocess your images to resize them to (64, 64) and convert them to grayscale before feeding them into the model:

   from tensorflow.keras.preprocessing.image import ImageDataGenerator

   datagen = ImageDataGenerator(rescale=1./255)
   train_generator = datagen.flow_from_directory(
       'path/to/dataset',
       target_size=(64, 64),  # Resize images to 64x64
       color_mode='grayscale',  # Convert images to grayscale
       class_mode='categorical'
   )

   from tensorflow.keras.preprocessing.image import ImageDataGenerator

   datagen = ImageDataGenerator(rescale=1./255)
   train_generator = datagen.flow_from_directory(
       'path/to/dataset',
       target_size=(64, 64),  # Resize images to 64x64
       color_mode='grayscale',  # Convert images to grayscale
       class_mode='categorical'
   )

2. If you prefer to use the images in their original size and color, you’ll need to adjust the model’s input shape accordingly:

   model = models.Sequential([
       layers.Conv2D(16, (3, 3), activation='relu', input_shape=(128, 128, 3)),
       layers.MaxPooling2D((2, 2), name="pool_1"),
       layers.Conv2D(32, (3, 3), activation='relu', name="conv_2"),
       layers.MaxPooling2D((2, 2), name="pool_2"),
       layers.Flatten(name="flatten"),
       layers.Dense(64, activation='relu', name="dense_1"),
       layers.Dense(3, activation='softmax', name="output_layer")
   ])

   model = models.Sequential([
       layers.Conv2D(16, (3, 3), activation='relu', input_shape=(128, 128, 3)),
       layers.MaxPooling2D((2, 2), name="pool_1"),
       layers.Conv2D(32, (3, 3), activation='relu', name="conv_2"),
       layers.MaxPooling2D((2, 2), name="pool_2"),
       layers.Flatten(name="flatten"),
       layers.Dense(64, activation='relu', name="dense_1"),
       layers.Dense(3, activation='softmax', name="output_layer")
   ])

Enthernet Code•3mo ago

@RED HAT Thanks for the detailed response, I see the issue now. I'll try the first option since memory usage is a big concern for this project. Resizing the images and converting them to grayscale should keep things efficient.If I run into any other issues I’ll reach out. I appreciate the help 👍

Gaming

Programming

How do I fix a tensor dimension mismatch in TinyML disease detection?