Need a set of test images (keras.cifar.load_data() only has train/val)
Consider using Bing’s image generator to create new images for testing and/or ask pax to bring their own
27.07 EG Found a new dataset CINIC-10 with same format as CIFAR that can be divided into train, val, and test sets (cf keras.loadcifar() returns only train, val)
EG Note dataset is 670MB to download and can expand to >1GB depending on disk format. It’s 90K images in three sets (270K total) and takes my tf-GPU ~20min to load. Need to reconsider how to use.
Need a set of test images (keras.cifar.load_data() only has train/val)
Consider using Bing’s image generator to create new images for testing and/or ask pax to bring their own
27.07 EG Found a new dataset CINIC-10 with same format as CIFAR that can be divided into train, val, and test sets (cf keras.loadcifar() returns only train, val)
EG Note dataset is 670MB to download and can expand to >1GB depending on disk format. It’s 90K images in three sets (270K total) and takes my tf-GPU ~20min to load. Need to reconsider how to use.