How to choose the size of the input layer in CNN for object detection?

Hi all,

I have collected a dataset for object detection the size of the images are 480x360. I want to know how to define the size of the input layer on my CNN so that I get the best result. does the input size layer influence the accuracy of my CNN?

Thanks in advance

Answer 1

[Meet]

Hi Sehairi,

From my understanding of the question, you want to know the best size of the input layer to the CNN and does the input size layer influence the accuracy of the CNN.

Since your images are 480x360, you might consider using this size directly as the input to preserve the original aspect ratio and resolution. This helps in maintaining the spatial relationships in the image, which can be important factor for object detection.

Yes, the size of the input layer does affect the accuracy of the CNN model, larger input sizes can capture more detail, potentially improving accuracy, especially for small objects. However, they also increase computation time. Conversely, smaller input sizes process faster but may lose important details. I would suggest experimenting with different input sizes to find the best balance between accuracy and performance of the model.

You can refer to this article for more detailed information:

https://www.researchgate.net/publication/332241609_IMPACT_OF_IMAGE_SIZE_ON_ACCURACY_AND_GENERALIZATION_OF_CONVOLUTIONAL_NEURAL_NETWORKS

I hope this information is helpful.