Enhancing the accuracy of object recognition model on smart edge devices
DOI:
https://doi.org/10.54654/isj.v2i19.948Keywords:
DL model, edge device, real time detection, object detectionTóm tắt
Abstract—Object recognition is one of the main topics in the AI field. There are many AI models with high accuracy running well on high-configuration devices. However, smart edge devices (SED) are being widely used in many different fields because of their compact flexibility, ensuring personal data policy. Their limitation is hardware that only runs or supports 8bits 16bits or 32bits models. Therefore, running the model on SED must do the swap step (“quantization”). This also causes the recognition models to be significantly reduced in accuracy. In this paper, we propose the solution “GreedyPlus” – to capture high resolution frame (skip blur frame) and search for small objects in the image by cutting frames into small windows. Then, the solution zooms in and identifies objects. The last step determines the number of objects in the frame exactly. The method is simple but highly effective, improving the recognition results for the model significantly without the need to retrain the model with a new dataset. The results are tested and demonstrated on the datasets KITTI, CrownAI, and Autti.
Downloads
References
. Kim, Jangho, KiYoon Yoo, and Nojun Kwak. "Position-based scaled gradient for model quantization and pruning." Advances in Neural Information Processing Systems 33 (2020): 20415-20426.
. Li, Rundong, et al. "Fully quantized network for object detection." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019.
. Krishnamoorthi, Raghuraman. "Quantizing deep convolutional networks for efficient inference: A whitepaper." arXiv preprint arXiv:1806.08342 (2018).
. Ren, Shaoqing, et al. "Faster r-cnn: Towards real-time object detection with region proposal networks." Advances in neural information processing systems 28 (2015).
. Dai, Jifeng, Yi Li, Kaiming He, and Jian Sun. "R-fcn: Object detection via region-based fully convolutional networks." Advances in neural information processing systems 29 (2016).
. Lin, Tsung-Yi, et al. "Focal loss for dense object detection." Proceedings of the IEEE international conference on computer vision. 2017.
. Redmon, Joseph, and Ali Farhadi. "Yolov3: An incremental improvement." arXiv preprint arXiv:1804.02767 (2018).
. Liu, Wei, et al. "Ssd: Single shot multibox detector." Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14. Springer International Publishing, 2016.
. Sandler, Mark, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. "Mobilenetv2: Inverted residuals and linear bottlenecks." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4510-4520. 2018.
. Howard, Andrew G., Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. "Mobilenets: Efficient convolutional neural networks for mobile vision applications." arXiv preprint arXiv:1704.04861 (2017).
. Thuật toán Non maximum suppression https://learnopencv.com/non-maximum-suppression-theory-and-implementation-in-pytorch/.
Downloads
Published
How to Cite
Issue
Section
License
Proposed Policy for Journals That Offer Open Access
Authors who publish with this journal agree to the following terms:
1. Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
2. Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
3. Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).
Proposed Policy for Journals That Offer Delayed Open Access
Authors who publish with this journal agree to the following terms:
1. Authors retain copyright and grant the journal right of first publication, with the work [SPECIFY PERIOD OF TIME] after publication simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
2. Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
3. Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).