Hi everyone,
I am working with a high-quality image dataset, pixel size (720*1280).
I am trying to train a CNN model for classification using this dataset but the computation takes a long time, what techniques that I can use to reduce the computational complexity?
Thanks in advance.
You cannot reduce complexity but you could
1. Try some library that can use GPU. For example Tensorflow (in essence that is parallel computing as José Arzola-Ruiz suggested)
2. Resizing the pictures should reduce the needed time, so experiment with it and see if it its of some use.
I hope that this helps.
There can be many ways to reduce computations in CNNs.
1. Reduce the image size which will reduce the feature size hence computations
2. Reduce the number of layers/nodes per layer this can also cause underfitting/overfitting but depends on the size of the dataset.
3. Use CUDA based GPU, thus not reducing the computations but enhancing the computational capability (Machine Level Parallelism).
4. Use skip connections concept
5. Use a more normalised dataset, be careful do not alter the features (significantly) while normalising.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training