The Scalable, Adaptive Localization, and Laplacian Object Proposal (Scallop) Toolkit is a small computer vision toolkit aimed at detecting roughly elliptical or blob-like objects in imagery. It is brief port of some of my master's thesis code developed at RPI, described further in the paper:

Automatic Scallop Detection in Benthic Environments

Beyond what's in the paper, the toolkit has a few more modern optimizations as well, such as the ability to run convolutional neural networks on top of the described object proposal framework. It is useful as a general detector for detecting any ellipsoidal objects, though it also contains specialized subroutines targetting benthic organisms such as clams, scallops, urchins, and others.

WARNING: This repository is still a little bit of a WIP because I haven't touched the code in a few years and it was created when I was a student. Code quality varies greatly file to file.

Requirements: CMake, OpenCV, Caffe (Optional)

First, install CMake and build or install OpenCV and Caffe.

Next, checkout this repository, run CMake on it, point CMake to the installed dependencies, and then build using your compiler of choice.

Alternatively, ScallopTK can be built in VIAME via enabling it in the build settings (set VIAME_ENABLE_CAFFE=ON and VIAME_ENABLE_SCALLOP_TK=ON). This can be useful and easier since it also builds all of the dependencies of caffe.

I also recommend installing CUDA >= 7.0 if you have an NVIDIA graphics card for the computational speed-up, prior to building Caffe.

A manual is provided, though it is very out of date and in need of updating. The most basic way to run the core detector pipeline is to run:

./ScallopDetector PROCESS_DIR InputDirectoryWithImages OutputDetectionFile.txt

You can switch between AdaBoost, CNN, and Combo classifiers in the SYSTEM_SETTINGS file.