Training ultrasound images with original CNN architectures directly is not satisfactory for object detection, segmentation, etc. Is it possible to improve CNN performance for these images?
Generally the CNN architectures used for segmentation etc. captures the variation using the convolutional filters present in the model and gives better segmentation even without improving the resolution. If they are not working properly, try to use some techniques to first enhance the resolution (using some super-resolution techniques) and then use those images as input to train the network combined with the original raw images.
if your test and train set of similar resolution then a good CNN that can learn the features should work. If you think than your CNN not working because of the low-resolution images, then first you need to check other factors , are you having a balanced data, is the CNN deep enough to learn what you want to learn...etc.
Dear Navchetan Awasthi and Reem Alaasam Thank you for your answers.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training