Classifier-Free Guidance (CFG) enhances the quality and condition adherence of text-to-image diffusion models. It operates by combining the conditional and unconditional predictions using a fixed weight. However, recent works vary the weights throughout the diffusion process, reporting superior results but without providing any rationale or analysis. By conducting comprehensive experiments, this paper provides insights into CFG weight schedulers. Our findings suggest that simple, monotonically increasing weight schedulers consistently lead to improved performances, requiring merely a single line of code. In addition, more complex parametrized schedulers can be optimized for further improvement, but do not generalize across different models and tasks.
In this paper we propose a new non-linear classifier based on a combination of locally linear classifiers. A well known optimization formulation is given as we cast the problem in a $\ell_1$ Multiple Kernel Learning (MKL) problem using many locally linear kernels. Since the number of such kernels is huge, we provide a scalable generic MKL training algorithm handling streaming kernels. With respect to the inference time, the resulting classifier fits the gap between high accuracy but slow non-linear classifiers (such as classical MKL) and fast but low accuracy linear classifiers.
In recent research, Learned Image Compression has gained prominence for its capacity to outperform traditional handcrafted pipelines, especially at low bit-rates. While existing methods incorporate convolutional priors with occasional attention blocks to address long-range dependencies, recent advances in computer vision advocate for a transformative shift towards fully transformer-based architectures grounded in the attention mechanism. This paper investigates the feasibility of image compression exclusively using attention layers within our novel model, QPressFormer. We introduce the concept of learned image queries to aggregate patch information via cross-attention, followed by quantization and coding techniques. Through extensive evaluations, our work demonstrates competitive performance achieved by convolution-free architectures across the popular Kodak, DIV2K, and CLIC datasets.
Class-Incremental Learning (CIL) aims to build classification models from data streams. At each step of the CIL process, new classes must be integrated into the model. Due to catastrophic forgetting, CIL is particularly challenging when examples from past classes cannot be stored, the case on which we focus here. To date, most approaches are based exclusively on the target dataset of the CIL process. However, the use of models pre-trained in a self-supervised way on large amounts of data has recently gained momentum. The initial model of the CIL process may only use the first batch of the target dataset, or also use pre-trained weights obtained on an auxiliary dataset. The choice between these two initial learning strategies can significantly influence the performance of the incremental learning model, but has not yet been studied in depth. Performance is also influenced by the choice of the CIL algorithm, the neural architecture, the nature of the target task, the distribution of classes in the stream and the number of examples available for learning. We conduct a comprehensive experimental study to assess the roles of these factors. We present a statistical analysis framework that quantifies the relative contribution of each factor to incremental performance. Our main finding is that the initial training strategy is the dominant factor influencing the average incremental accuracy, but that the choice of CIL algorithm is more important in preventing forgetting. Based on this analysis, we propose practical recommendations for choosing the right initial training strategy for a given incremental learning use case. These recommendations are intended to facilitate the practical deployment of incremental learning.
This paper presents a new approach to image similarity search in the context of fashion, a domain with inherent ambiguity due to the multiple ways in which images can be considered similar. We introduce the concept of Referred Visual Search (RVS), where users provide additional information to define the desired similarity. We present a new dataset, LAION-RVS-Fashion, consisting of 272K fashion products with 842K images extracted from LAION, designed explicitly for this task. We then propose an innovative method for learning conditional embeddings using weakly-supervised training, achieving a 6% increase in Recall at one (R@1) against a gallery with 2M distractors, compared to classical approaches based on explicit attention and filtering. The proposed method demonstrates robustness, maintaining similar R@1 when dealing with 2.5 times as many distractors as the baseline methods. We believe this is a step forward in the emerging field of Referred Visual Search both in terms of accessible data and approach. Code, data and models are available at https://www.github.com/Simon-Lepage/CondViT-LRVSF .
Solving jigsaw puzzles requires to grasp the visual features of a sequence of patches and to explore efficiently a solution space that grows exponentially with the sequence length. Therefore, visual deep reinforcement learning (DRL) should answer this problem more efficiently than optimization solvers coupled with neural networks. Based on this assumption, we introduce Alphazzle, a reassembly algorithm based on single-player Monte Carlo Tree Search (MCTS). A major difference with DRL algorithms lies in the unavailability of game reward for MCTS, and we show how to estimate it from the visual input with neural networks. This constraint is induced by the puzzle-solving task and dramatically adds to the task complexity (and interest!). We perform an in-deep ablation study that shows the importance of MCTS and the neural networks working together. We achieve excellent results and get exciting insights into the combination of DRL and visual feature learning.
Recent advances in visual representation learning allowed to build an abundance of powerful off-the-shelf features that are ready-to-use for numerous downstream tasks. This work aims to assess how well these features preserve information about the objects, such as their spatial location, their visual properties and their relative relationships. We propose to do so by evaluating them in the context of visual reasoning, where multiple objects with complex relationships and different attributes are at play. More specifically, we introduce a protocol to evaluate visual representations for the task of Visual Question Answering. In order to decouple visual feature extraction from reasoning, we design a specific attention-based reasoning module which is trained on the frozen visual representations to be evaluated, in a spirit similar to standard feature evaluations relying on shallow networks. We compare two types of visual representations, densely extracted local features and object-centric ones, against the performances of a perfect image representation using ground truth. Our main findings are two-fold. First, despite excellent performances on classical proxy tasks, such representations fall short for solving complex reasoning problem. Second, object-centric features better preserve the critical information necessary to perform visual reasoning. In our proposed framework we show how to methodologically approach this evaluation.
3D human whole-body pose estimation aims to localize precise 3D keypoints on the entire human body, including the face, hands, body, and feet. Due to the lack of a large-scale fully annotated 3D whole-body dataset, a common approach has been to train several deep networks separately on datasets dedicated to specific body parts, and combine them during inference. This approach suffers from complex training and inference pipelines because of the different biases in each dataset used. It also lacks a common benchmark which makes it difficult to compare different methods. To address these issues, we introduce Human3.6M 3D WholeBody (H3WB) which provides whole-body annotations for the Human3.6M dataset using the COCO Wholebody layout. H3WB is a large scale dataset with 133 whole-body keypoint annotations on 100K images, made possible by our new multi-view pipeline. Along with H3WB, we propose 3 tasks: i) 3D whole-body pose lifting from 2D complete whole-body pose, ii) 3D whole-body pose lifting from 2D incomplete whole-body pose, iii) 3D whole-body pose estimation from a single RGB image. We also report several baselines from popular methods for these tasks. The dataset is publicly available at \url{https://github.com/wholebody3d/wholebody3d}.
Exemplar-free class-incremental learning is very challenging due to the negative effect of catastrophic forgetting. A balance between stability and plasticity of the incremental process is needed in order to obtain good accuracy for past as well as new classes. Existing exemplar-free class-incremental methods focus either on successive fine tuning of the model, thus favoring plasticity, or on using a feature extractor fixed after the initial incremental state, thus favoring stability. We introduce a method which combines a fixed feature extractor and a pseudo-features generator to improve the stability-plasticity balance. The generator uses a simple yet effective geometric translation of new class features to create representations of past classes, made of pseudo-features. The translation of features only requires the storage of the centroid representations of past classes to produce their pseudo-features. Actual features of new classes and pseudo-features of past classes are fed into a linear classifier which is trained incrementally to discriminate between all classes. The incremental process is much faster with the proposed method compared to mainstream ones which update the entire deep model. Experiments are performed with three challenging datasets, and different incremental settings. A comparison with ten existing methods shows that our method outperforms the others in most cases.
A large body of recent work targets semantically conditioned image generation. Most such methods focus on the narrower task of pose transfer and ignore the more challenging task of subject transfer that consists in not only transferring the pose but also the appearance and background. In this work, we introduce SCAM (Semantic Cross Attention Modulation), a system that encodes rich and diverse information in each semantic region of the image (including foreground and background), thus achieving precise generation with emphasis on fine details. This is enabled by the Semantic Attention Transformer Encoder that extracts multiple latent vectors for each semantic region, and the corresponding generator that exploits these multiple latents by using semantic cross attention modulation. It is trained only using a reconstruction setup, while subject transfer is performed at test time. Our analysis shows that our proposed architecture is successful at encoding the diversity of appearance in each semantic region. Extensive experiments on the iDesigner and CelebAMask-HD datasets show that SCAM outperforms SEAN and SPADE; moreover, it sets the new state of the art on subject transfer.