GenRecal: Generation after Recalibration from Large to Small Vision-Language Models

Recent advancements in vision-language models (VLMs) have leveraged large language models (LLMs) to achieve performance on par with closed-source systems like GPT-4V. However, deploying these models in real-world scenarios, particularly on resource-constrained devices, remains challenging due to their substantial computational demands. This has spurred interest in distilling knowledge from large VLMs into smaller, more efficient counterparts. A key challenge arises here from the diversity of VLM architectures, which are built on different LLMs and employ varying token types-differing in vocabulary size, token splits, and token index ordering. To address this challenge of limitation to a specific VLM type, we present Generation after Recalibration (GenRecal), a novel, general-purpose distillation framework for VLMs. GenRecal incorporates a Recalibrator that aligns and adapts feature representations between heterogeneous VLMs, enabling effective knowledge transfer across different types of VLMs. Through extensive experiments on multiple challenging benchmarks, we demonstrate that GenRecal significantly improves baseline performances, eventually outperforming large-scale open- and closed-source VLMs.

Omni-RGPT: Unifying Image and Video Region-level Understanding via Token Marks

We present Omni-RGPT, a multimodal large language model designed to facilitate region-level comprehension for both images and videos. To achieve consistent region representation across spatio-temporal dimensions, we introduce Token Mark, a set of tokens highlighting the target regions within the visual feature space. These tokens are directly embedded into spatial regions using region prompts (e.g., boxes or masks) and simultaneously incorporated into the text prompt to specify the target, establishing a direct connection between visual and text tokens. To further support robust video understanding without requiring tracklets, we introduce an auxiliary task that guides Token Mark by leveraging the consistency of the tokens, enabling stable region interpretation across the video. Additionally, we introduce a large-scale region-level video instruction dataset (RegVID-300k). Omni-RGPT achieves state-of-the-art results on image and video-based commonsense reasoning benchmarks while showing strong performance in captioning and referring expression comprehension tasks.

VLsI: Verbalized Layers-to-Interactions from Large to Small Vision Language Models

The recent surge in high-quality visual instruction tuning samples from closed-source vision-language models (VLMs) such as GPT-4V has accelerated the release of open-source VLMs across various model sizes. However, scaling VLMs to improve performance using larger models brings significant computational challenges, especially for deployment on resource-constrained devices like mobile platforms and robots. To address this, we propose VLsI: Verbalized Layers-to-Interactions, a new VLM family in 2B and 7B model sizes, which prioritizes efficiency without compromising accuracy. VLsI leverages a unique, layer-wise distillation process, introducing intermediate "verbalizers" that map features from each layer to natural language space, allowing smaller VLMs to flexibly align with the reasoning processes of larger VLMs. This approach mitigates the training instability often encountered in output imitation and goes beyond typical final-layer tuning by aligning the small VLMs' layer-wise progression with that of the large ones. We validate VLsI across ten challenging vision-language benchmarks, achieving notable performance gains (11.0% for 2B and 17.4% for 7B) over GPT-4V without the need for model scaling, merging, or architectural changes.

RealTraj: Towards Real-World Pedestrian Trajectory Forecasting

This paper jointly addresses three key limitations in conventional pedestrian trajectory forecasting: pedestrian perception errors, real-world data collection costs, and person ID annotation costs. We propose a novel framework, RealTraj, that enhances the real-world applicability of trajectory forecasting. Our approach includes two training phases--self-supervised pretraining on synthetic data and weakly-supervised fine-tuning with limited real-world data--to minimize data collection efforts. To improve robustness to real-world errors, we focus on both model design and training objectives. Specifically, we present Det2TrajFormer, a trajectory forecasting model that remains invariant in tracking noise by using past detections as inputs. Additionally, we pretrain the model using multiple pretext tasks, which enhance robustness and improve forecasting performance based solely on detection data. Unlike previous trajectory forecasting methods, our approach fine-tunes the model using only ground-truth detections, significantly reducing the need for costly person ID annotations. In the experiments, we comprehensively verify the effectiveness of the proposed method against the limitations, and the method outperforms state-of-the-art trajectory forecasting methods on multiple datasets.

SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP

Large-scale vision-language models, such as CLIP, are known to contain harmful societal bias regarding protected attributes (e.g., gender and age). In this paper, we aim to address the problems of societal bias in CLIP. Although previous studies have proposed to debias societal bias through adversarial learning or test-time projecting, our comprehensive study of these works identifies two critical limitations: 1) loss of attribute information when it is explicitly disclosed in the input and 2) use of the attribute annotations during debiasing process. To mitigate societal bias in CLIP and overcome these limitations simultaneously, we introduce a simple-yet-effective debiasing method called SANER (societal attribute neutralizer) that eliminates attribute information from CLIP text features only of attribute-neutral descriptions. Experimental results show that SANER, which does not require attribute annotations and preserves original information for attribute-specific descriptions, demonstrates superior debiasing ability than the existing methods.

CrowdMAC: Masked Crowd Density Completion for Robust Crowd Density Forecasting

A crowd density forecasting task aims to predict how the crowd density map will change in the future from observed past crowd density maps. However, the past crowd density maps are often incomplete due to the miss-detection of pedestrians, and it is crucial to develop a robust crowd density forecasting model against the miss-detection. This paper presents a MAsked crowd density Completion framework for crowd density forecasting (CrowdMAC), which is simultaneously trained to forecast future crowd density maps from partially masked past crowd density maps (i.e., forecasting maps from past maps with miss-detection) while reconstructing the masked observation maps (i.e., imputing past maps with miss-detection). Additionally, we propose Temporal-Density-aware Masking (TDM), which non-uniformly masks tokens in the observed crowd density map, considering the sparsity of the crowd density maps and the informativeness of the subsequent frames for the forecasting task. Moreover, we introduce multi-task masking to enhance training efficiency. In the experiments, CrowdMAC achieves state-of-the-art performance on seven large-scale datasets, including SDD, ETH-UCY, inD, JRDB, VSCrowd, FDST, and croHD. We also demonstrate the robustness of the proposed method against both synthetic and realistic miss-detections.

From Descriptive Richness to Bias: Unveiling the Dark Side of Generative Image Caption Enrichment

Large language models (LLMs) have enhanced the capacity of vision-language models to caption visual text. This generative approach to image caption enrichment further makes textual captions more descriptive, improving alignment with the visual context. However, while many studies focus on benefits of generative caption enrichment (GCE), are there any negative side effects? We compare standard-format captions and recent GCE processes from the perspectives of "gender bias" and "hallucination", showing that enriched captions suffer from increased gender bias and hallucination. Furthermore, models trained on these enriched captions amplify gender bias by an average of 30.9% and increase hallucination by 59.5%. This study serves as a caution against the trend of making captions more descriptive.

Multimodal Cross-Domain Few-Shot Learning for Egocentric Action Recognition

We address a novel cross-domain few-shot learning task (CD-FSL) with multimodal input and unlabeled target data for egocentric action recognition. This paper simultaneously tackles two critical challenges associated with egocentric action recognition in CD-FSL settings: (1) the extreme domain gap in egocentric videos (\eg, daily life vs. industrial domain) and (2) the computational cost for real-world applications. We propose MM-CDFSL, a domain-adaptive and computationally efficient approach designed to enhance adaptability to the target domain and improve inference speed. To address the first challenge, we propose the incorporation of multimodal distillation into the student RGB model using teacher models. Each teacher model is trained independently on source and target data for its respective modality. Leveraging only unlabeled target data during multimodal distillation enhances the student model's adaptability to the target domain. We further introduce ensemble masked inference, a technique that reduces the number of input tokens through masking. In this approach, ensemble prediction mitigates the performance degradation caused by masking, effectively addressing the second issue. Our approach outperformed the state-of-the-art CD-FSL approaches with a substantial margin on multiple egocentric datasets, improving by an average of 6.12/6.10 points for 1-shot/5-shot settings while achieving $2.2$ times faster inference speed. Project page: https://masashi-hatano.github.io/MM-CDFSL/

EMAG: Ego-motion Aware and Generalizable 2D Hand Forecasting from Egocentric Videos

Predicting future human behavior from egocentric videos is a challenging but critical task for human intention understanding. Existing methods for forecasting 2D hand positions rely on visual representations and mainly focus on hand-object interactions. In this paper, we investigate the hand forecasting task and tackle two significant issues that persist in the existing methods: (1) 2D hand positions in future frames are severely affected by ego-motions in egocentric videos; (2) prediction based on visual information tends to overfit to background or scene textures, posing a challenge for generalization on novel scenes or human behaviors. To solve the aforementioned problems, we propose EMAG, an ego-motion-aware and generalizable 2D hand forecasting method. In response to the first problem, we propose a method that considers ego-motion, represented by a sequence of homography matrices of two consecutive frames. We further leverage modalities such as optical flow, trajectories of hands and interacting objects, and ego-motions, thereby alleviating the second issue. Extensive experiments on two large-scale egocentric video datasets, Ego4D and EPIC-Kitchens 55, verify the effectiveness of the proposed method. In particular, our model outperforms prior methods by $7.0$\% on cross-dataset evaluations. Project page: https://masashi-hatano.github.io/EMAG/

Weakly Semi-supervised Tool Detection in Minimally Invasive Surgery Videos

Surgical tool detection is essential for analyzing and evaluating minimally invasive surgery videos. Current approaches are mostly based on supervised methods that require large, fully instance-level labels (i.e., bounding boxes). However, large image datasets with instance-level labels are often limited because of the burden of annotation. Thus, surgical tool detection is important when providing image-level labels instead of instance-level labels since image-level annotations are considerably more time-efficient than instance-level annotations. In this work, we propose to strike a balance between the extremely costly annotation burden and detection performance. We further propose a co-occurrence loss, which considers a characteristic that some tool pairs often co-occur together in an image to leverage image-level labels. Encapsulating the knowledge of co-occurrence using the co-occurrence loss helps to overcome the difficulty in classification that originates from the fact that some tools have similar shapes and textures. Extensive experiments conducted on the Endovis2018 dataset in various data settings show the effectiveness of our method.