AntoineSalaun / PlanktonTaxaClassifier

Testing new technics to improve Plankton taxonomic classification such as ResNets and Discriminative filters DFL-VGG16.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

PlanktonTaxaClassifier

Download the dataset

The dataset canbe downloaded from SEANOE's website : https://www.seanoe.org/data/00446/55741/ Then, move the data in the ZooScanSet folder

Start a run

You can start a run by launching the following command

cd Code
python3 Runner.py 1.0 0.7 20 1e-5 128 ADAM 0.1 ResNet18

Training parameters

The parameters should come after

python3 Runner.py

in the following order

  • sampling factor (What percentage of the data is used)
  • train factor (How much of it is used for training, the rest is for validation)
  • epochs
  • learning rate
  • Batch Size
  • Optimizer (ADAM or SGD)
  • Weight decay
  • Network (BaseNet, ResNet18, ResNet34 or DFL-VGG16)

Note

Relative paths are used but depending on the architecture of your file system, some paths might be broken. If it is the case, you should modify it in the beggining of Code/Experiment.py

About

Testing new technics to improve Plankton taxonomic classification such as ResNets and Discriminative filters DFL-VGG16.


Languages

Language:Python 98.3%Language:Rich Text Format 1.7%