Evaluating Text-to-Image Synthesis: Survey and Taxonomy of Image Quality Metrics

Recent advances in text-to-image synthesis enabled through a combination of language and vision foundation models have led to a proliferation of the tools available and an increased attention to the field. When conducting text-to-image synthesis, a central goal is to ensure that the content between text and image is aligned. As such, there exist numerous evaluation metrics that aim to mimic human judgement. However, it is often unclear which metric to use for evaluating text-to-image synthesis systems as their evaluation is highly nuanced. In this work, we provide a comprehensive overview of existing text-to-image evaluation metrics. Based on our findings, we propose a new taxonomy for categorizing these metrics. Our taxonomy is grounded in the assumption that there are two main quality criteria, namely compositionality and generality, which ideally map to human preferences. Ultimately, we derive guidelines for practitioners conducting text-to-image evaluation, discuss open challenges of evaluation mechanisms, and surface limitations of current metrics.

Evaluating Text to Image Synthesis: Survey and Taxonomy of Image Quality Metrics

Recent advances in text-to-image synthesis have been enabled by exploiting a combination of language and vision through foundation models. These models are pre-trained on tremendous amounts of text-image pairs sourced from the World Wide Web or other large-scale databases. As the demand for high-quality image generation shifts towards ensuring content alignment between text and image, novel evaluation metrics have been developed with the aim of mimicking human judgments. Thus, researchers have started to collect datasets with increasingly complex annotations to study the compositionality of vision-language models and their incorporation as a quality measure of compositional alignment between text and image contents. In this work, we provide a comprehensive overview of existing text-to-image evaluation metrics and propose a new taxonomy for categorizing these metrics. We also review frequently adopted text-image benchmark datasets before discussing techniques to optimize text-to-image synthesis models towards quality and human preferences. Ultimately, we derive guidelines for improving text-to-image evaluation and discuss the open challenges and current limitations.

TTT-KD: Test-Time Training for 3D Semantic Segmentation through Knowledge Distillation from Foundation Models

Test-Time Training (TTT) proposes to adapt a pre-trained network to changing data distributions on-the-fly. In this work, we propose the first TTT method for 3D semantic segmentation, TTT-KD, which models Knowledge Distillation (KD) from foundation models (e.g. DINOv2) as a self-supervised objective for adaptation to distribution shifts at test-time. Given access to paired image-pointcloud (2D-3D) data, we first optimize a 3D segmentation backbone for the main task of semantic segmentation using the pointclouds and the task of 2D $\to$ 3D KD by using an off-the-shelf 2D pre-trained foundation model. At test-time, our TTT-KD updates the 3D segmentation backbone for each test sample, by using the self-supervised task of knowledge distillation, before performing the final prediction. Extensive evaluations on multiple indoor and outdoor 3D segmentation benchmarks show the utility of TTT-KD, as it improves performance for both in-distribution (ID) and out-of-distribution (ODO) test datasets. We achieve a gain of up to 13% mIoU (7% on average) when the train and test distributions are similar and up to 45% (20% on average) when adapting to OOD test samples.

Attention-Guided Masked Autoencoders For Learning Image Representations

Masked autoencoders (MAEs) have established themselves as a powerful method for unsupervised pre-training for computer vision tasks. While vanilla MAEs put equal emphasis on reconstructing the individual parts of the image, we propose to inform the reconstruction process through an attention-guided loss function. By leveraging advances in unsupervised object discovery, we obtain an attention map of the scene which we employ in the loss function to put increased emphasis on reconstructing relevant objects, thus effectively incentivizing the model to learn more object-focused representations without compromising the established masking strategy. Our evaluations show that our pre-trained models learn better latent representations than the vanilla MAE, demonstrated by improved linear probing and k-NN classification results on several benchmarks while at the same time making ViTs more robust against varying backgrounds.

Spatially Guiding Unsupervised Semantic Segmentation Through Depth-Informed Feature Distillation and Sampling

Traditionally, training neural networks to perform semantic segmentation required expensive human-made annotations. But more recently, advances in the field of unsupervised learning have made significant progress on this issue and towards closing the gap to supervised algorithms. To achieve this, semantic knowledge is distilled by learning to correlate randomly sampled features from images across an entire dataset. In this work, we build upon these advances by incorporating information about the structure of the scene into the training process through the use of depth information. We achieve this by (1) learning depth-feature correlation by spatially correlate the feature maps with the depth maps to induce knowledge about the structure of the scene and (2) implementing farthest-point sampling to more effectively select relevant features by utilizing 3D sampling techniques on depth information of the scene. Finally, we demonstrate the effectiveness of our technical contributions through extensive experimentation and present significant improvements in performance across multiple benchmark datasets.

Leveraging Self-Supervised Vision Transformers for Neural Transfer Function Design

In volume rendering, transfer functions are used to classify structures of interest, and to assign optical properties such as color and opacity. They are commonly defined as 1D or 2D functions that map simple features to these optical properties. As the process of designing a transfer function is typically tedious and unintuitive, several approaches have been proposed for their interactive specification. In this paper, we present a novel method to define transfer functions for volume rendering by leveraging the feature extraction capabilities of self-supervised pre-trained vision transformers. To design a transfer function, users simply select the structures of interest in a slice viewer, and our method automatically selects similar structures based on the high-level features extracted by the neural network. Contrary to previous learning-based transfer function approaches, our method does not require training of models and allows for quick inference, enabling an interactive exploration of the volume data. Our approach reduces the amount of necessary annotations by interactively informing the user about the current classification, so they can focus on annotating the structures of interest that still require annotation. In practice, this allows users to design transfer functions within seconds, instead of minutes. We compare our method to existing learning-based approaches in terms of annotation and compute time, as well as with respect to segmentation accuracy. Our accompanying video showcases the interactivity and effectiveness of our method.