dmarx whats-in-a-name issues

dmarx / whats-in-a-name

[WIP] probing identity and bias in text to image models

MIT License

1 stars 1 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

other relevant techniques

#30 dmarx opened 7 months ago
0
estimate an empirical prior from LAION-Face

#29 dmarx opened 7 months ago
1
find a decent, lightweight face parsing model for face segmentation

#28 dmarx closed 8 months ago
5
ignore facial hair in skin tone mask segmentation

#27 dmarx opened 8 months ago
0
names scrape

#26 dmarx opened 8 months ago
2
seed-constrained identity strength measure

#25 dmarx opened 9 months ago
1
nearest neighbors basis!

#24 dmarx opened 9 months ago
2
better embedding visualization

#23 dmarx opened 9 months ago
0
play w captum methods

#22 dmarx opened 9 months ago
0
Ethnology of facial phenotypes in CLIP

#21 dmarx opened 9 months ago
2
investigate (model's) global biases by prompting for "person from <country>"

#20 dmarx opened 9 months ago
3
experimetn with arize-ai/phoenix alternative embedding explorer

#19 dmarx opened 9 months ago
1
scrape some ground truth images

#18 dmarx opened 9 months ago
0
use TAD to bin embeddings into model prior vs. strongly distinct identities

#17 dmarx opened 9 months ago
0
crop faces

#16 dmarx opened 9 months ago
1
measure strength/flexibility per-attribute

#15 dmarx opened 9 months ago
2
CKA variety of prompts for same name to elucidate where in network identity-relevant bits are

#14 dmarx opened 9 months ago
1
fancy embeddings visualizaiton

#13 dmarx closed 9 months ago
0
misc relevant papers

#12 dmarx opened 9 months ago
5
grokking stuff

#11 dmarx opened 9 months ago
0
attempt to discover relevant neural circuits - mechanistic interpretability

#10 dmarx opened 9 months ago
0
Try DIV metric from dreambooth

#9 dmarx closed 9 months ago
1
Reproduce paper: Disentanglement of Correlated Factors via Hausdorff Factorized Support

#8 dmarx opened 10 months ago
1
textual inversion + PEZ = prompt-space personalization

#7 dmarx opened 10 months ago
2
use latent structural conditioing to promote "portrait" images

#6 dmarx closed 9 months ago
2
does focusing on high/(low?)-strength identity generations make it easier/harder to fine tune a model?

#5 dmarx opened 10 months ago
0
Compare across CLIP checkpoints

#4 dmarx opened 10 months ago
1
does identity strength of init token impact personalization performance (LoRA dreambooth)

#3 dmarx opened 10 months ago
0
CLIP+DIP to probe extent to which these identities are endemic to the CLIP model directly

#2 dmarx opened 10 months ago
0
control for seed diversity

#1 dmarx opened 10 months ago
1