Sort images in a folder based on similarity features. The default featurizer is 'mobilenet+resnet18' hybrid, which performs the best combining OD and Classification features.
python sort.py /path/to/input/directory/ /path/to/output/directory/
To use different featurizer
python sort.py /path/to/input/directory /path/to/output/directory/ --featurizer mobilenet
Decrease the dedup threshold to deduplicate more
python sort.py /path/to/input/directory /path/to/output/directory/ --dedup 0.99
- Paralelize and cache to disk to handle large number of images. (Currently fine up to few hundreds)
- Add option to minimize the cost function (max sum path) for maximum similarity sort. Can start from current sort as a good initial condition.