Skip to content

NeuonAI/plantclef2021_challenge

Repository files navigation

Neuon AI in PlantCLEF 2021

This repository contains the team's work in PlantCLEF 2021. The goal of the challenge was to identify plants in field pictures based on a training set of digitized herbarium specimens. We achieved an MRR of 0.181 on the primary metric and an MRR of 0.158 on the secondary metric (difficult species). The official results are released here.

Methodology

Stage 1: Construct and train networks

Herbarium-Field Triplet Loss Network (HFTL Network)

Figure 1

One-streamed Network (OSM Network)

Figure 2

Stage 2: Obtain herbarium dictionary

Herbarium Dictionary Construction

Figure 3

Stage 3: Compare feature similarity between field embedding and herbarium dictionary

Feature similarity comparison

Figure 4

Repository files

Training scripts

Validation scripts

Lists

Checkpoints

About

Training lists and scripts used in PlantCLEF 2021

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages