Commit 3d5fb1b6 authored by Amelie Royer's avatar Amelie Royer 🐼

Upload New File

parent 08edfd05
- Data examples: colorized ILSVRC and spatial ILSVRC. Different visual domains but same target task
- setting = only have a pretrained model on data X, and limited amound of annotated samples from data Y
- baseline = fine-tuning later layers, keeping the early one frozen
- proposed: "projecting" data Y to data X at the feature level
- fine-tuning
require a layer criterion selection = analyze accuracy when replacing early layers by ones finetuned on the other dataset
select the "turning-point layer" as the one with the highest drop in accurcy
~ similar to criterion in How "transferable are features in deep neural networks?", Yosinski et al.
- projection module: image-to-image module. Mainly useful for parametriwed transformations
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment