Clustering

[Zoyela]

• [x] try out clustering with triplet loss
• [ ] try out magnetic loss if it worked good and there is enough time

Images:

• 224x224 with resnet18
• Output 128
• same learning rate as for the resnet50

Folder structure for the clustering

• Root/Train/Cam12_morning_noAnimal
• Root/Train/Cam12_morning_animal
• ....

[SebastianLukas]

There should be results this night late. Folder structures is: CamA_Empty_Day, CamA_Empty_Night, CamA_Empty_DawnOrDusk, CamA_NotEmpty_Day, etc. For each camera. I did exclude Classes for which i got less than 50 images

[Zoyela]

Try out clustering with images of 3 cameras: Put all the images together and let nn cluster them into surrogate classes. Then train and test a resnet to classify images into one of the classes defined through the clustering. Paper: Unsupervised Feature Learning for Writer Identification and Writer Retrieval