2 research outputs found
Making Images Speak: Human-Inspired Image Description Generation
No full textDespite significant advances in deep learning-based image captioning, many state-of-the-art models still struggle to balance visual grounding (i.e., accurate object and scene descriptions) with linguistic coherence (i.e., grammatical fluency and appropriate use of non-visual tokens such as articles and prepositions). To address these limitations, we propose a hybrid image captioning framework that integrates handcrafted and deep visual features. Specifically, we combine local descriptors—Scale-Invariant Feature Transform (SIFT) and Bag of Features (BoF)—with high-level semantic features extracted using ResNet50. This dual representation captures both fine-grained spatial details and contextual semantics. The decoder employs Bahdanau attention refined with an Attention-on-Attention (AoA) mechanism to optimize visual-textual alignment, while GloVe embeddings and a GRU-based sequence model ensure fluent language generation. The proposed system is trained on 200,000 image-caption pairs from the MS COCO train2014 dataset and evaluated on 50,000 held-out MS COCO pairs plus the Flickr8K benchmark. Our model achieves a CIDEr score of 128.3 and a SPICE score of 29.24, reflecting clear improvements over baselines in both semantic precision—particularly for spatial relationships—and grammatical fluency. These results validate that combining classical computer vision techniques with modern attention mechanisms yields more interpretable and linguistically precise captions, addressing key limitations in neural caption generation
Broadband and Low-Frequency Acoustic Liner Investigations at NASA and ONERA
No full textInternational audienceNASA and ONERA have explored a number of acoustic liner concepts over the last few decades. This paper begins with a brief review regarding conventional liners as well as the recent implementation of multi-degree-of-freedom liners enabled by embedded mesh caps. Six novel liner concepts are presented, along with the accompanying impedance prediction models used in their design. Each of the NASA concepts is designed to vary the impedance over the surface of the liner in a controlled manner, whereas the ONERA concepts make use of long neck acoustic resonators. Selected results are presented for each of these liners evaluated in the NASA and ONERA test rigs. Finally, a set of aeroacoustic metrics is defined for comparison standardization between conventional and innovative acoustic liners and all the concepts are compared on this basis
