Neuon AI in PlantCLEF 2020

This repository contains the team's work in PlantCLEF 2020.

We constructed a two-streamed Herbarium-Field Triplet Loss Network in tackling the challenge. The network is illustrated below:

Repository Files

Training scripts

• run_inception_triplet_loss.py
• database_module_triplet_loss.py

Validation scripts

• get_997_herbarium_mean_emb_crop.py
  • gets the herbarium embedding representation for each class
• validate_image.py
  • validate a test image

Lists

• clef2020_herbarium_species.csv
  • list of species mapped to trained index
• clef2020_herbarium_species_classid_map_to_index.txt
  • list of class id (folder) mapped to trained index
• clef2020_known_classes_herbarium_train.txt
  • herbarium network training dataset
• clef2020_known_classes_herbarium_train_added.txt
  • herbarium network training dataset with added herbarium test dataset
• clef2020_known_classes_herbarium_test.txt
  • herbarium network test dataset
• clef2020_known_classes_field_train.txt
  • field network training dataset
• clef2020_known_classes_field_train_added.txt
  • field network training dataset with added field test dataset
• clef2020_known_classes_field_test.txt
  • field network test dataset
• clef2020_multilabel_train.txt
  • list of herbarium used in get_997_herbarium_mean_emb_crop.py

Checkpoint

• 004000.ckpt.7z.001
  • chekpoint of the trained network