search

samuelstevens / swin-transformer

This is an official implementation for "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows".
https://arxiv.org/abs/2103.14030
MIT License
0 stars 0 forks source link

iNat21 (funky-banana) #9

Closed samuelstevens closed 1 year ago

samuelstevens commented 1 year ago

Pre-train swin-v2-base on iNat21 with default settings for 90 epochs at 192x192, then 30 epochs at 256x256.

samuelstevens commented 1 year ago

I am tracking this experiment in funky-banana.

samuelstevens commented 1 year ago

This experiment is strictly worse than training with a LR of 1.25e-4, so while I am letting it run because I don't need the GPU hours right now, I don't expect to ever use this checkpoint.