Visual Context Window Extension: A New Perspective for Long Video Understanding

Large Multimodal Models (LMMs) have demonstrated impressive performance in short video understanding tasks but face great challenges when applied to long video understanding. In contrast, Large Language Models (LLMs) exhibit outstanding capabilities in modeling long texts. Existing work attempts to address this issue by introducing long video-text pairs during training. However, these approaches require substantial computational and data resources. In this paper, we tackle the challenge of long video understanding from the perspective of context windows, aiming to apply LMMs to long video tasks without retraining on long video datasets. We first conduct an in-depth analysis of why pretrained LMMs struggle to understand lengthy video content, identifying that discrepancies between visual and language modalities lead to different context windows for visual and language tokens, making it difficult to directly extend the visual tokens to match the language context window. Based on this, we propose to adapt LMMs for long video understanding tasks by extending the visual context window, eliminating the need for retraining on large scalelong video datasets. To further mitigate the significant memory consumption caused by long sequences, we introduce a progressive pooling inference strategy that selectively adjusts the spatial resolution of frame embeddings, reducing the number of visual tokens while retaining important spatial information. Across multiple long video understanding benchmarks, our method consistently improves the performance as the number of video frames increases. On the MLVU benchmark, our method outperforms GPT-4o, even though our model size is only 7B. Additionally, in the 256-frame setting, our method reduces memory usage by approximately 45% compared to the baseline, without introducing any performance loss.

Improving Generalization of Image Captioning with Unsupervised Prompt Learning

Pretrained visual-language models have demonstrated impressive zero-shot abilities in image captioning, when accompanied by hand-crafted prompts. Meanwhile, hand-crafted prompts utilize human prior knowledge to guide the model. However, due to the diversity between different domains, such hand-crafted prompt that provide invariant prior knowledge may result in mode collapse for some domains. Some researches attempted to incorporate expert knowledge and instruction datasets, but the results were costly and led to hallucinations. In this paper, we propose an unsupervised prompt learning method to improve Generalization of Image Captioning (GeneIC), which learns a domain-specific prompt vector for the target domain without requiring annotated data. GeneIC aligns visual and language modalities with a pre-trained Contrastive Language-Image Pre-Training (CLIP) model, thus optimizing the domain-specific prompt vector from two aspects: attribute and semantic consistency. Specifically, GeneIC first generates attribute-transferred images with differing attributes, while retaining semantic similarity with original images. Then, GeneIC uses CLIP to measure the similarity between the images and the generated sentences. By exploring the variable and invariant features in the original images and attribute-transferred images, attribute consistency constrains the attribute change direction of both images and sentences to learn domain-specific knowledge. The semantic consistency directly measures the similarity between the generated sentences and images to ensure the accuracy and comprehensiveness of the generated sentences. Consequently, GeneIC only optimizes the prompt vectors, which effectively retains the knowledge in the large model and introduces domain-specific knowledge.

Exploiting Cross-Modal Prediction and Relation Consistency for Semi-Supervised Image Captioning

The task of image captioning aims to generate captions directly from images via the automatically learned cross-modal generator. To build a well-performing generator, existing approaches usually need a large number of described images, which requires a huge effects on manual labeling. However, in real-world applications, a more general scenario is that we only have limited amount of described images and a large number of undescribed images. Therefore, a resulting challenge is how to effectively combine the undescribed images into the learning of cross-modal generator. To solve this problem, we propose a novel image captioning method by exploiting the Cross-modal Prediction and Relation Consistency (CPRC), which aims to utilize the raw image input to constrain the generated sentence in the commonly semantic space. In detail, considering that the heterogeneous gap between modalities always leads to the supervision difficulty of using the global embedding directly, CPRC turns to transform both the raw image and corresponding generated sentence into the shared semantic space, and measure the generated sentence from two aspects: 1) Prediction consistency. CPRC utilizes the prediction of raw image as soft label to distill useful supervision for the generated sentence, rather than employing the traditional pseudo labeling; 2) Relation consistency. CPRC develops a novel relation consistency between augmented images and corresponding generated sentences to retain the important relational knowledge. In result, CPRC supervises the generated sentence from both the informativeness and representativeness perspectives, and can reasonably use the undescribed images to learn a more effective generator under the semi-supervised scenario.