Existing video analysis models often lack explainability, perform poorly on long videos, and frequently hallucinate. Commercial solutions are closed-source and costly. We introduce CReLeRI, an open-source system for action detection in untrimmed videos. CReLeRI segments videos using scene and action transitions, detects actions and their arguments and grounds them in 3D space to improve interpretability and reduce hallucinations. The system promotes transparency and trust in AI-driven analysis of complex, real-world videos. A demonstration video is also available.

In this paper, a novel early forest fire detection and localization system is proposed based on an unmanned aerial vehicle (UAV) equipped with RGB and thermal cameras. The forest fire detection is achieved by the proposed bimodal detection network. Fire localization, i.e., fire distance estimation, is realized by the proposed monocular depth estimation network and scale recovery strategy. To validate the effectiveness of the proposed method, the performance of the detection network is tested using the bimodal dataset UAVFire presented in this paper and the public dataset FLAME2. The proposed depth estimation network is validated on the public dataset WildUAV. The experimental results show that the proposed bimodal detection network has a AP of 97.9% and 98.2% on the UAVFire and FLAME2 datasets, respectively, with a model parameter of 6.5 MB. Compared with other unimodal detection methods as well as bimodal detection methods, the detection method in this paper performs well in terms of both the detection accuracy and the model lightweighting. The model parameter of the proposed monocular depth estimation network is only 0.69 MB, and the comparison results with other methods on theWildUAV dataset prove that the depth estimation method in this paper achieves a good balance between accuracy and model lightweighting. Finally, the detection and distance estimation of early small fires are validated on the dataset UAVFire, demonstrating the effectiveness of the proposed method.

This paper presents a lightweight method for detecting forest fires using dual-modal remote sensing images captured by an unmanned aerial vehicle (UAV). The aim is to achieve efficient fire monitoring on a computationally resource-constrained UAV platform. The proposed detection network is based on the improved YOLOv8, which uses RGB image and thermal image as network input at the same time. A lightweight dual-modal feature fusion module named DFM is designed to effectively combine RGB and thermal features. The existing C2f module in YOLOv8 was replaced by the lightweight module C2f-F, along with the addition of the parameter-free attention module SimAM. This improvement improves the detection performance of the model while minimizing the model parameters. The evaluation experimental results on the FLAME 2 dataset show that the accuracy of the proposed dual-modal forest fire detection method reaches 98.4%, and the model size is only 2.9MB, which achieves a good balance between accuracy and number of parameters compared with other mainstream methods. Additionally, on the iCrest 2-s edge computing device, the detection speed reaches 20.67FPS, further confirming that this lightweight approach satisfies the real-time detection requirements for forest fires.

Many text generation applications require the generated text to be factually consistent with input information.Automatic evaluation of factual consistency is challenging.Previous work has developed various metrics that often depend on specific functions, such as natural language inference (NLI) or question answering (QA), trained on limited data.Those metrics thus can hardly assess diverse factual inconsistencies (e.g., contradictions, hallucinations) that occur in varying inputs/outputs (e.g., sentences, documents) from different tasks.In this paper, we propose ALIGNSCORE, a new holistic metric that applies to a variety of factual inconsistency scenarios as above.ALIGN-SCORE is based on a general function of information alignment between two arbitrary text pieces.Crucially, we develop a unified training framework of the alignment function by integrating a large diversity of data sources, resulting in 4.7M training examples from 7 well-established tasks (NLI, QA, paraphrasing, fact verification, information retrieval, semantic similarity, and summarization).We conduct extensive experiments on large-scale benchmarks including 22 evaluation datasets, where 19 of the datasets were never seen in the alignment training.ALIGNSCORE achieves substantial improvement over a wide range of previous metrics.Moreover, ALIGNSCORE (355M parameters) matches or even outperforms metrics based on ChatGPT and GPT-4 that are orders of magnitude larger. 1

Large language models (LLMs), typically designed as a function of next-word prediction, have excelled across extensive NLP tasks. Despite the generality, next-word prediction is often not an efficient formulation for many of the tasks, demanding an extreme scale of model parameters (10s or 100s of billions) and sometimes yielding suboptimal performance. In practice, it is often desirable to build more efficient models---despite being less versatile, they still apply to a substantial subset of problems, delivering on par or even superior performance with much smaller model sizes. In this paper, we propose text alignment as an efficient unified model for a wide range of crucial tasks involving text entailment, similarity, question answering (and answerability), factual consistency, and so forth. Given a pair of texts, the model measures the degree of alignment between their information. We instantiate an alignment model through lightweight finetuning of RoBERTa (355M parameters) using 5.9M examples from 28 datasets. Despite its compact size, extensive experiments show the model's efficiency and strong performance: (1) On over 20 datasets of aforementioned diverse tasks, the model matches or surpasses FLAN-T5 models that have around 2x or 10x more parameters; the single unified model also outperforms task-specific models finetuned on individual datasets; (2) When applied to evaluate factual consistency of language generation on 23 datasets, our model improves over various baselines, including the much larger GPT-3.5 (ChatGPT) and sometimes even GPT-4; (3) The lightweight model can also serve as an add-on component for LLMs such as GPT-3.5 in question answering tasks, improving the average exact match (EM) score by 17.94 and F1 score by 15.05 through identifying unanswerable questions.