For predicates whose usage and parameters are not obvious.
DNN + JIT SVM: Binary classification. Just-in-time train an SVM using the provided zip file as training set. The zip file should contain a positive/ folder and a negative/ folder. The input feature to the SVM is extracted using a pre-trained DNN (e.g., MobileNet trained on ImageNet).
DNN ImageNet Classify: Classification. Use a pre-trained DNN to classify the images into the 1000 ImageNet classes. You should give labels from those 1000 class names.
SS + DNN + JIT SVM: Object detection. Use selective search as the region proposal algorithm. Then chose a small number of regions and run DNN + JIT SVM on them.
TPOD Wrapper: Object detection. Wrap an object detector created using TPOD (A Tool for Painless Object Detection) developed at CMU.
SIFT/SURF homography: Key point matching/object detection. Match the images with the provided example patches using SIFT or SURF. The filter will pass an image if at least one example patch matches.