search

AMI-system / gbif-species-trainer-AMI-fork

Code for training a fine-grained species classification model using data from GBIF
MIT License
0 stars 0 forks source link

replicate the updated Rolnick model #2

Closed KatrionaGoldmann closed 11 months ago

KatrionaGoldmann commented 1 year ago

Add pytorch based model to model sub-directory

LevanBokeria commented 1 year ago

Updated the issue name to reflect the goal of replicating the Rolnick model exactly, with the same dataset they mention in READMEs.

KatrionaGoldmann commented 1 year ago

Currently running using CPUs on baskerville, but then running into memory and/or time capacity issues.

New issue opened: Run models on baskerville with GPU/CUDA#17 to resolve improve the time capacity barrier

KatrionaGoldmann commented 1 year ago

Much faster now using GPUs. Model is predicted to run to completion in around 2.5 hours, however we run into disk quota issues.

KatrionaGoldmann commented 1 year ago

Model running in https://github.com/AMI-trap/on_device_classifier/pull/19

KatrionaGoldmann commented 11 months ago

We can achieve similar accuracies using the increased number of images (aim for 1000 per species).

Models run for:

See https://github.com/AMI-trap/on_device_classifier/commit/6919a39cdbf900407dc605414067769a76084c04