Can We Use Saliency Maps for the Evaluation of Generative Image Models?

Generative models can produce images nearly indistinguishable from real data, yet rigorous and interpretable evaluation remains challenging. Conventional metrics such as FID provide only scalar scores with limited diagnostic insight. Widely adopted CLIP-based metrics enable semantic evaluation beyond simple training class labels, but inherit limitations from CLIP's training paradigm that restrict attribute-wise analysis. We propose RA-CLIPScore, a novel metric that mitigates these issues and extends CLIP-based evaluation to spatial distribution alignment, measuring whether generated objects adhere to the positional priors found in the training data. RA-CLIPScore introduces dual prompts to decouple competing attributes and leverages local patch tokens to capture fine-grained regional semantics. We evaluate image generative models on their ability to match both attribute and spatial distributions of the training data. Extensive experiments show that RA-CLIPScore provides more robust and interpretable evaluations than prior methods, particularly under distribution misalignment or partially irrelevant textual attributes. We further demonstrate how it reveals spatial biases in generative models. User evaluations confirm that Regional Single Attribute Divergence based on our RA-CLIPScore aligns more closely with human perception of visual diversity than existing semantic metrics.

This paper studies the backbone used by metrics for evaluating the quality of generated images. Although generative models have advanced rapidly, some outputs still exhibit unrealistic texture characteristics, and these differences are difficult to measure. Widely used evaluation metrics provide limited insight into texture realism, since they often rely on a backbone trained for another objective, such as image classification or image-text alignment. Such objectives typically encourage invariance to low-level variations that carry texture, so the resulting features primarily encode semantic and global structure. As a result, texture-specific discrepancies may not be explicitly captured. We propose TRIG, a metric directed at the texture realism of generated images. Its backbone is instead the discriminator of a generative adversarial network pre-trained on a texture dataset, which provides a learned prior for the local statistics of natural textures. A plausibility head is fine-tuned to identify local texture degradations, while preserving natural defocus and sensor noise as positive cases. On the features of this backbone, we compute a standard distributional distance between a reference set of real images and a set of generated ones. We apply TRIG to several generative models and observe closer agreement with human realness ratings than baseline metrics.

The rapid proliferation of generative models raises the model attribution problem: given only an image, can we determine which model produced it? We propose a lightweight CNN to solve this problem in a strict black box setting. The CNN operates on multiple image patches to handle varying image size and improve accuracy. It attributes more models at higher accuracy than prior work, reaching 98.9% on 25-class DRAGON and 95.0% on 27-class OpenFake; runs in a few milliseconds at a cost nearly independent of candidate-set size; and remains robust to transformations encountered in the wild. Beyond closed-set attribution, its learned representation serves as a reusable fingerprint backbone: it supports open-set rejection and few-shot enrollment, recovers model-family structure without lineage supervision, and groups images from unseen generators by source. Controlled ablations and causal perturbations show that its attribution decisions are driven by a spatially local, low-level signal that behaves as a generator-specific fingerprint.