ChatSearch: a Dataset and a Generative Retrieval Model for General Conversational Image Retrieval

In this paper, we investigate the task of general conversational image retrieval on open-domain images. The objective is to search for images based on interactive conversations between humans and computers. To advance this task, we curate a dataset called ChatSearch. This dataset includes a multi-round multimodal conversational context query for each target image, thereby requiring the retrieval system to find the accurate image from database. Simultaneously, we propose a generative retrieval model named ChatSearcher, which is trained end-to-end to accept/produce interleaved image-text inputs/outputs. ChatSearcher exhibits strong capability in reasoning with multimodal context and can leverage world knowledge to yield visual retrieval results. It demonstrates superior performance on the ChatSearch dataset and also achieves competitive results on other image retrieval tasks and visual conversation tasks. We anticipate that this work will inspire further research on interactive multimodal retrieval systems. Our dataset will be available at https://github.com/joez17/ChatSearch.

SubZero: Random Subspace Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning

Fine-tuning Large Language Models (LLMs) has proven effective for a variety of downstream tasks. However, as LLMs grow in size, the memory demands for backpropagation become increasingly prohibitive. Zeroth-order (ZO) optimization methods offer a memory-efficient alternative by using forward passes to estimate gradients, but the variance of gradient estimates typically scales linearly with the model's parameter dimension$\unicode{x2013}$a significant issue for LLMs. In this paper, we propose the random Subspace Zeroth-order (SubZero) optimization to address the challenges posed by LLMs' high dimensionality. We introduce a low-rank perturbation tailored for LLMs that significantly reduces memory consumption while improving training performance. Additionally, we prove that our gradient estimation closely approximates the backpropagation gradient, exhibits lower variance than traditional ZO methods, and ensures convergence when combined with SGD. Experimental results show that SubZero enhances fine-tuning performance and achieves faster convergence compared to standard ZO approaches like MeZO across various language modeling tasks.

Unifying Event-based Flow, Stereo and Depth Estimation via Feature Similarity Matching

As an emerging vision sensor, the event camera has gained popularity in various vision tasks such as optical flow estimation, stereo matching, and depth estimation due to its high-speed, sparse, and asynchronous event streams. Unlike traditional approaches that use specialized architectures for each specific task, we propose a unified framework, EventMatch, that reformulates these tasks as an event-based dense correspondence matching problem, allowing them to be solved with a single model by directly comparing feature similarities. By utilizing a shared feature similarities module, which integrates knowledge from other event flows via temporal or spatial interactions, and distinct task heads, our network can concurrently perform optical flow estimation from temporal inputs (e.g., two segments of event streams in the temporal domain) and stereo matching from spatial inputs (e.g., two segments of event streams from different viewpoints in the spatial domain). Moreover, we further demonstrate that our unified model inherently supports cross-task transfer since the architecture and parameters are shared across tasks. Without the need for retraining on each task, our model can effectively handle both optical flow and disparity estimation simultaneously. The experiment conducted on the DSEC benchmark demonstrates that our model exhibits superior performance in both optical flow and disparity estimation tasks, outperforming existing state-of-the-art methods. Our unified approach not only advances event-based models but also opens new possibilities for cross-task transfer and inter-task fusion in both spatial and temporal dimensions. Our code will be available later.

Temporal Residual Guided Diffusion Framework for Event-Driven Video Reconstruction

Event-based video reconstruction has garnered increasing attention due to its advantages, such as high dynamic range and rapid motion capture capabilities. However, current methods often prioritize the extraction of temporal information from continuous event flow, leading to an overemphasis on low-frequency texture features in the scene, resulting in over-smoothing and blurry artifacts. Addressing this challenge necessitates the integration of conditional information, encompassing temporal features, low-frequency texture, and high-frequency events, to guide the Denoising Diffusion Probabilistic Model (DDPM) in producing accurate and natural outputs. To tackle this issue, we introduce a novel approach, the Temporal Residual Guided Diffusion Framework, which effectively leverages both temporal and frequency-based event priors. Our framework incorporates three key conditioning modules: a pre-trained low-frequency intensity estimation module, a temporal recurrent encoder module, and an attention-based high-frequency prior enhancement module. In order to capture temporal scene variations from the events at the current moment, we employ a temporal-domain residual image as the target for the diffusion model. Through the combination of these three conditioning paths and the temporal residual framework, our framework excels in reconstructing high-quality videos from event flow, mitigating issues such as artifacts and over-smoothing commonly observed in previous approaches. Extensive experiments conducted on multiple benchmark datasets validate the superior performance of our framework compared to prior event-based reconstruction methods.

Beyond Image Prior: Embedding Noise Prior into Conditional Denoising Transformer

Existing learning-based denoising methods typically train models to generalize the image prior from large-scale datasets, suffering from the variability in noise distributions encountered in real-world scenarios. In this work, we propose a new perspective on the denoising challenge by highlighting the distinct separation between noise and image priors. This insight forms the basis for our development of conditional optimization framework, designed to overcome the constraints of traditional denoising framework. To this end, we introduce a Locally Noise Prior Estimation (LoNPE) algorithm, which accurately estimates the noise prior directly from a single raw noisy image. This estimation acts as an explicit prior representation of the camera sensor's imaging environment, distinct from the image prior of scenes. Additionally, we design an auxiliary learnable LoNPE network tailored for practical application to sRGB noisy images. Leveraging the estimated noise prior, we present a novel Conditional Denoising Transformer (Condformer), by incorporating the noise prior into a conditional self-attention mechanism. This integration allows the Condformer to segment the optimization process into multiple explicit subspaces, significantly enhancing the model's generalization and flexibility. Extensive experimental evaluations on both synthetic and real-world datasets, demonstrate that the proposed method achieves superior performance over current state-of-the-art methods. The source code is available at https://github.com/YuanfeiHuang/Condformer.

Exploiting Frequency Correlation for Hyperspectral Image Reconstruction

Deep priors have emerged as potent methods in hyperspectral image (HSI) reconstruction. While most methods emphasize space-domain learning using image space priors like non-local similarity, frequency-domain learning using image frequency priors remains neglected, limiting the reconstruction capability of networks. In this paper, we first propose a Hyperspectral Frequency Correlation (HFC) prior rooted in in-depth statistical frequency analyses of existent HSI datasets. Leveraging the HFC prior, we subsequently establish the frequency domain learning composed of a Spectral-wise self-Attention of Frequency (SAF) and a Spectral-spatial Interaction of Frequency (SIF) targeting low-frequency and high-frequency components, respectively. The outputs of SAF and SIF are adaptively merged by a learnable gating filter, thus achieving a thorough exploitation of image frequency priors. Integrating the frequency domain learning and the existing space domain learning, we finally develop the Correlation-driven Mixing Domains Transformer (CMDT) for HSI reconstruction. Extensive experiments highlight that our method surpasses various state-of-the-art (SOTA) methods in reconstruction quality and computational efficiency.

4-bit Shampoo for Memory-Efficient Network Training

Second-order optimizers, maintaining a matrix termed a preconditioner, are superior to first-order optimizers in both theory and practice. The states forming the preconditioner and its inverse root restrict the maximum size of models trained by second-order optimizers. To address this, compressing 32-bit optimizer states to lower bitwidths has shown promise in reducing memory usage. However, current approaches only pertain to first-order optimizers. In this paper, we propose the first 4-bit second-order optimizers, exemplified by 4-bit Shampoo, maintaining performance similar to that of 32-bit ones. We show that quantizing the eigenvector matrix of the preconditioner in 4-bit Shampoo is remarkably better than quantizing the preconditioner itself both theoretically and experimentally. By rectifying the orthogonality of the quantized eigenvector matrix, we enhance the approximation of the preconditioner's eigenvector matrix, which also benefits the computation of its inverse 4-th root. Besides, we find that linear square quantization slightly outperforms dynamic tree quantization when quantizing second-order optimizer states. Evaluation on various networks for image classification demonstrates that our 4-bit Shampoo achieves comparable test accuracy to its 32-bit counterpart while being more memory-efficient. The source code will be made available.

MixLight: Borrowing the Best of both Spherical Harmonics and Gaussian Models

Accurately estimating scene lighting is critical for applications such as mixed reality. Existing works estimate illumination by generating illumination maps or regressing illumination parameters. However, the method of generating illumination maps has poor generalization performance and parametric models such as Spherical Harmonic (SH) and Spherical Gaussian (SG) fall short in capturing high-frequency or low-frequency components. This paper presents MixLight, a joint model that utilizes the complementary characteristics of SH and SG to achieve a more complete illumination representation, which uses SH and SG to capture low-frequency ambient and high-frequency light sources respectively. In addition, a special spherical light source sparsemax (SLSparsemax) module that refers to the position and brightness relationship between spherical light sources is designed to improve their sparsity, which is significant but omitted by prior works. Extensive experiments demonstrate that MixLight surpasses state-of-the-art (SOTA) methods on multiple metrics. In addition, experiments on Web Dataset also show that MixLight as a parametric method has better generalization performance than non-parametric methods.

SpikeNeRF: Learning Neural Radiance Fields from Continuous Spike Stream

Spike cameras, leveraging spike-based integration sampling and high temporal resolution, offer distinct advantages over standard cameras. However, existing approaches reliant on spike cameras often assume optimal illumination, a condition frequently unmet in real-world scenarios. To address this, we introduce SpikeNeRF, the first work that derives a NeRF-based volumetric scene representation from spike camera data. Our approach leverages NeRF's multi-view consistency to establish robust self-supervision, effectively eliminating erroneous measurements and uncovering coherent structures within exceedingly noisy input amidst diverse real-world illumination scenarios. The framework comprises two core elements: a spike generation model incorporating an integrate-and-fire neuron layer and parameters accounting for non-idealities, such as threshold variation, and a spike rendering loss capable of generalizing across varying illumination conditions. We describe how to effectively optimize neural radiance fields to render photorealistic novel views from the novel continuous spike stream, demonstrating advantages over other vision sensors in certain scenes. Empirical evaluations conducted on both real and novel realistically simulated sequences affirm the efficacy of our methodology. The dataset and source code are released at https://github.com/BIT-Vision/SpikeNeRF.

Finding Visual Saliency in Continuous Spike Stream

As a bio-inspired vision sensor, the spike camera emulates the operational principles of the fovea, a compact retinal region, by employing spike discharges to encode the accumulation of per-pixel luminance intensity. Leveraging its high temporal resolution and bio-inspired neuromorphic design, the spike camera holds significant promise for advancing computer vision applications. Saliency detection mimics the behavior of human beings and captures the most salient region from the scenes. In this paper, we investigate the visual saliency in the continuous spike stream for the first time. To effectively process the binary spike stream, we propose a Recurrent Spiking Transformer (RST) framework, which is based on a full spiking neural network. Our framework enables the extraction of spatio-temporal features from the continuous spatio-temporal spike stream while maintaining low power consumption. To facilitate the training and validation of our proposed model, we build a comprehensive real-world spike-based visual saliency dataset, enriched with numerous light conditions. Extensive experiments demonstrate the superior performance of our Recurrent Spiking Transformer framework in comparison to other spike neural network-based methods. Our framework exhibits a substantial margin of improvement in capturing and highlighting visual saliency in the spike stream, which not only provides a new perspective for spike-based saliency segmentation but also shows a new paradigm for full SNN-based transformer models. The code and dataset are available at \url{https://github.com/BIT-Vision/SVS}.