The Caffe Model Zoo (http://caffe.berkeleyvision.org/model_zoo.html) provides various pre-trained DNN models that I would like to use. However, I'm struggling to find the list of classes and objects that these nets were trained on and their corresponding indices.
Does anyone know where to find this? I've looked into the readme files and the actual protobuf but found nothing.
Thanks!
The models in the model zoo are usually trained/tested on a selected few publicly available benchmarks. I believe each model states clearly what benchmark it was trained/tested on. Knowing what benchmark was used you can access the labels list from the benchmark website directly.
For instance, models trained on ImageNet (aka ILSVRC challenge) are trained to classify each image into one of 1000 classes (last layer's num_output: 1000). You can find more information about the classes of this benchmark here and here.
Some models are trained on CIFAR tiny images. This benchmark has two "flavors": CIFAR10 and CIFAR100 that classify each image into one of 10 or 100 classes respectively (num_output of last layer either 10 or 100). You can find a list of the classes on the benchmark's webpage.
Apart from classification models there are detection models, trained to classify and localize multiple objects in each image. These models are usually trained on benchmarks such as MSCOCO which has 80 classes (if I'm not mistaken), and PASCAL VOC detection challenge which has 20 classes.
Bottom line, you need to see for each model on what benchmark it was trained and then take the labels from the benchmark's webpage.
