Large Multimodal Models (LMMs) have shown significant reasoning capabilities by connecting a visual encoder and a large language model. LMMs typically use a fixed amount of visual tokens, such as the penultimate layer features in the CLIP visual encoder, as the prefix content. Recent LMMs incorporate more complex visual inputs, such as high-resolution images and videos, which increase the number of visual tokens significantly. However, due to the design of the Transformer architecture, computational costs associated with these models tend to increase quadratically with the number of input tokens. To tackle this problem, we explore a token reduction mechanism and find, similar to prior work, that many visual tokens are spatially redundant. Based on this, we propose PruMerge, a novel adaptive visual token reduction approach, which largely reduces the number of visual tokens while maintaining comparable model performance. We first select the unpruned visual tokens based on their similarity to class tokens and spatial tokens. We then cluster the pruned tokens based on key similarity and merge the clustered tokens with the unpruned tokens to supplement their information. Empirically, when applied to LLaVA-1.5, our approach can compress the visual tokens by 18 times on average, and achieve comparable performance across diverse visual question-answering and reasoning tasks. Code and checkpoints are at https://llava-prumerge.github.io/.
The field of efficient Large Language Model (LLM) inference is rapidly evolving, presenting a unique blend of opportunities and challenges. Although the field has expanded and is vibrant, there hasn't been a concise framework that analyzes the various methods of LLM Inference to provide a clear understanding of this domain. Our survey stands out from traditional literature reviews by not only summarizing the current state of research but also by introducing a framework based on roofline model for systematic analysis of LLM inference techniques. This framework identifies the bottlenecks when deploying LLMs on hardware devices and provides a clear understanding of practical problems, such as why LLMs are memory-bound, how much memory and computation they need, and how to choose the right hardware. We systematically collate the latest advancements in efficient LLM inference, covering crucial areas such as model compression (e.g., Knowledge Distillation and Quantization), algorithm improvements (e.g., Early Exit and Mixture-of-Expert), and both hardware and system-level enhancements. Our survey stands out by analyzing these methods with roofline model, helping us understand their impact on memory access and computation. This distinctive approach not only showcases the current research landscape but also delivers valuable insights for practical implementation, positioning our work as an indispensable resource for researchers new to the field as well as for those seeking to deepen their understanding of efficient LLM deployment. The analyze tool, LLM-Viewer, is open-sourced.
Due to the lack of depth cues in images, multi-frame inputs are important for the success of vision-based perception, prediction, and planning in autonomous driving. Observations from different angles enable the recovery of 3D object states from 2D image inputs if we can identify the same instance in different input frames. However, the dynamic nature of autonomous driving scenes leads to significant changes in the appearance and shape of each instance captured by the camera at different time steps. To this end, we propose a novel contrastive learning algorithm, Cohere3D, to learn coherent instance representations in a long-term input sequence robust to the change in distance and perspective. The learned representation aids in instance-level correspondence across multiple input frames in downstream tasks. In the pretraining stage, the raw point clouds from LiDAR sensors are utilized to construct the long-term temporal correspondence for each instance, which serves as guidance for the extraction of instance-level representation from the vision-based bird's eye-view (BEV) feature map. Cohere3D encourages a consistent representation for the same instance at different frames but distinguishes between representations of different instances. We evaluate our algorithm by finetuning the pretrained model on various downstream perception, prediction, and planning tasks. Results show a notable improvement in both data efficiency and task performance.
We propose CounterCurate, a framework to comprehensively improve the visio-linguistic compositional reasoning capability for both contrastive and generative multimodal models. In particular, we identify two under-explored critical problems: the neglect of the physically grounded reasoning (counting and position understanding) and the potential of using highly capable text and image generation models for semantic counterfactual fine-tuning. Our work pioneers an approach that addresses these gaps. We first spotlight the near-chance performance of multimodal models like CLIP and LLaVA in physically grounded compositional reasoning. We then apply simple data augmentation using a grounded image generation model, GLIGEN, to generate finetuning data, resulting in significant performance improvements: +33% and +37% for CLIP and LLaVA, respectively, on our newly curated Flickr30k-Positions benchmark. Moreover, we exploit the capabilities of high-performing text generation and image generation models, specifically GPT-4V and DALLE-3, to curate challenging semantic counterfactuals, thereby further enhancing compositional reasoning capabilities on benchmarks such as SugarCrepe, where CounterCurate outperforms GPT-4V.
In recent years, image editing has advanced remarkably. With increased human control, it is now possible to edit an image in a plethora of ways; from specifying in text what we want to change, to straight up dragging the contents of the image in an interactive point-based manner. However, most of the focus has remained on editing single images at a time. Whether and how we can simultaneously edit large batches of images has remained understudied. With the goal of minimizing human supervision in the editing process, this paper presents a novel method for interactive batch image editing using StyleGAN as the medium. Given an edit specified by users in an example image (e.g., make the face frontal), our method can automatically transfer that edit to other test images, so that regardless of their initial state (pose), they all arrive at the same final state (e.g., all facing front). Extensive experiments demonstrate that edits performed using our method have similar visual quality to existing single-image-editing methods, while having more visual consistency and saving significant time and human effort.
We present FIND, a generalized interface for aligning foundation models' embeddings. As shown in teaser figure, a lightweight transformer interface without tuning any foundation model weights is enough for a unified image (segmentation) and dataset-level (retrieval) understanding. The proposed interface has the following favorable attributes: (1) Generalizable. It applies to various tasks spanning retrieval, segmentation, \textit{etc.}, under the same architecture and weights. (2) Prototypable. Different tasks are able to be implemented through prototyping attention masks and embedding types. (3) Extendable. The proposed interface is adaptive to new tasks, and new models. (4) Interleavable. With the benefit of multi-task multi-modal training, the proposed interface creates an interleaved shared embedding space. In light of the interleaved embedding space, we introduce the FIND-Bench, which introduces new training and evaluation annotations to the COCO dataset for interleave segmentation and retrieval. Our approach achieves state-of-the-art performance on FIND-Bench and competitive performance on standard retrieval and segmentation settings. The training, evaluation, and demo code as well as the dataset have been released at https://github.com/UX-Decoder/FIND.
Recent advances in generative deep learning have enabled the creation of high-quality synthetic images in text-to-image generation. Prior work shows that fine-tuning a pretrained diffusion model on ImageNet and generating synthetic training images from the finetuned model can enhance an ImageNet classifier's performance. However, performance degrades as synthetic images outnumber real ones. In this paper, we explore whether generative fine-tuning is essential for this improvement and whether it is possible to further scale up training using more synthetic data. We present a new framework leveraging off-the-shelf generative models to generate synthetic training images, addressing multiple challenges: class name ambiguity, lack of diversity in naive prompts, and domain shifts. Specifically, we leverage large language models (LLMs) and CLIP to resolve class name ambiguity. To diversify images, we propose contextualized diversification (CD) and stylized diversification (SD) methods, also prompted by LLMs. Finally, to mitigate domain shifts, we leverage domain adaptation techniques with auxiliary batch normalization for synthetic images. Our framework consistently enhances recognition model performance with more synthetic data, up to 6x of original ImageNet size showcasing the potential of synthetic data for improved recognition models and strong out-of-domain generalization.