MATH · IN · MODELS
methods / Causal Validation / Segmentation-scored channel identification + causal ablation/insertion

Segmentation-scored channel identification + causal ablation/insertion

Techniqueintermediate

Identifies individual feature-map channels ('units') whose spatial activation pattern aligns with a real object/concept segmentation mask (scored, e.g., by IoU across many images), then causally ablates or forces on the identified units and measures the resulting change in the network's own output.

Used in (1 observation)

structure: Linear Direction · models: Progressive GAN (trained on LSUN scene categories: bedroom, church, kitchen, etc.) · paper: GAN Dissection: Visualizing and Understanding Generative Adversarial Networks