Vision : Evaluate a real time image batch which is in memory


I have trained a model and now I want to predict on a test set. But this test set is generated dynamically in batches of 20. The images are in tensor format, in memory. I want to evaluate them without having to write them to files and then using the standard databunch api.

learn.predict(img) predicts on a single image. How can I evaluate on a batch of images together? Any help is much appreciated :slight_smile: