This repository contains notesbooks and code files which may be used to replicate the results from mil-benchmarks: Standardized Evaluation of Deep Multiple-Instance Learning Techniques.
| Notebook | Purpose |
|---|---|
| 00-mnist-generator | Generates Zero dataset |
| 01-fashion-generator | Generates Handbag dataset |
| 02-outfit-generator | Generates Outfit dataset |
| 03-mnist-cnn | Trains baseline model on Fashion and MNIST |
| 04-dense-mil | Trains Fully-Connect MIL model |
| 05a-max-pool-mil | Trains Max Pool MIL model |
| 05b-max-pool+dense-mil | Trains Max Pool + Fully-Connected MIL model |
| 07-noisy-and-mil | Trains Noisy-And MIL model |
| 08-noisy-data | Trains Noisy-And with label noise |