Skip to content
Data for knowledge distillation not always available due to copyrights and privacy concerns.Exist DFKD methods fail due to heavy reliance on BatchNorm layers → unusable in CLIP.Image-Text matching → DFKD for CLIP.
Inverse surrogate dataset from CLIP based on text prompts.Distill a student model from CLIP using surrogate dataset.Style dictionary diversification → diversity of synthetic images.Class consistency → prevent uncontrollable semantics introduced by diversification.
VLM
VLM
Paper Name
Categories
Type
Concept
Recent issues
Motivations
Contributions
Evaluation List
Target
Conf./Jour.
Year
Link
Open-Vocabulary Customization from CLIP via Data-Free Knowledge Distillation
Loss Function Design
There are no rows in this table
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.