Applications

Three following outside module are imported and repackaged in this repository, you are able to implement then by following the instruction in README.

tf_unet

tf_unet is a generic U-Net implementation as proposed by Ronneberger et al. developed with Tensorflow. In this repository, the repackaged module has following arguments:

• trainPath - the folder of training images
• testPath - the folder of testing images
• layerNum - the number of layers in the Unet, Suggestion: 3-5
• features - the number of features in the Unet
• bsize - training batch size, Suggestion: 2, 4
• opm - optimizer in Unet, default: adam
• iter - training iterations in one epoch of all data
• ep - the number of epoches in training
• display - the number of steps per display

thunder-extraction

thunder-extraction provides several algorithms including NMF that identifies spatial features of interest from data varying over space and time. In this repository, the repackaged module has following arguments:

• setName - the folder names of testing images
• base - where the files live, default: caesar server
• _k - k value in NMF
• _percentile - percentile value in NMF
• _max_iter - max iterations in NMF
• _overlap - overlap regions in NMF
• _chunk_size - chunk size in the process of NMF
• _padding - padding on the images in the process of NMF
• _merge - the number of regions to merge in NMF

CaImAn

CaImAn provides implementation of constrained NMF required in calcium imaging movies analysis pipeline. In this repository, the repackaged module has following arguments:

• setName - the folder names of testing images, default: all 9 testing sets
• k - K value in CNMF, number of neurons expected, default: k=1000
• g - gSig value in CNMF, expected half size of neurons, default: g=5
• merge - Merging threshold in CNMF, max correlation allowed, default: merge=0.8