Indoor–Outdoor Scene Classification with Residual Convolutional Neural Network
| Autor: | Seema Kumari, Arnav Bhavsar, Aditya Nigam, Ranjeet Ranjan Jha |
|---|---|
| Rok vydání: | 2019 |
| Předmět: | |
| Zdroj: | Proceedings of 3rd International Conference on Computer Vision and Image Processing ISBN: 9789813292901 CVIP (2) |
| DOI: | 10.1007/978-981-32-9291-8_26 |
| Popis: | In this paper, we demonstrate the effectiveness of a customized ResNet to address the problem of indoor–outdoor scene classification both for color images as well as depth images. Such an approach can serve as an initial step in a scene classification/retrieval pipeline or a single-image depth estimation task. The classification framework is developed based on Residual Convolutional Neural Network (ResNet-18) to classify any random scene as indoor or outdoor. We also demonstrate the invariance of the classification performance with respect to different weather conditions of outdoor scenes (which one can commonly encounter). The performance of our classification strategy is analyzed on different varieties of publicly available datasets of indoor and outdoor scenes that also have corresponding depth maps. The suggested approach achieves almost an ideal performance in many scenarios, for both color and depth images, across datasets. We also show positive comparisons with other state-of-the-art methods. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|