1. Company introduction
Define Co., Ltd. is a company that develops and supplies system software, develops and supplies optimal collection systems for household waste, eradicates harmful tides, and smart parking systems based on deep learning.
2. Problem Background and Summary
Although the service is being developed by applying a deep learning model that enables real-time object detection, performance delays are involved when performing image analysis on low-performance devices, and mathematical principles and improvement directions are derived to receive suggestions on how to change or lighten the model. The goal is to develop algorithms to detect harmful tides in low-performance devices.
3. Solving Process
Change settings for low-performance devices (Lotte Panda)
It has been confirmed that the actual detection speed can be improved without the need to modify or relearn the current model using input dimension changes and multi-core environment of the instrument, as well as the ability to reduce latency by adjusting the number of frames. Currently in preparation for field testing by applying this method directly to the prototype, Using SSD MOBILENET V1 Models
Using SSD MOBILENET, one of the most well-known deep learning techniques related to object detection, we have confirmed that the frame processing speed is higher than that of the existing use model, and we expect to get improved results from the actual low-performance device.
4. Ripple effects and future plans
If excellence is proven in the field, it is expected that economic wealth will be created not only for companies but also for the nation through expansion of regions across the country, expansion of overseas markets through diversification of products, and promotion of projects to link harmful tidal waves.
1. Company introduction
Define Co., Ltd. is a company that develops and supplies system software, develops and supplies optimal collection systems for household waste, eradicates harmful tides, and smart parking systems based on deep learning.
2. Problem Background and Summary
Although the service is being developed by applying a deep learning model that enables real-time object detection, performance delays are involved when performing image analysis on low-performance devices, and mathematical principles and improvement directions are derived to receive suggestions on how to change or lighten the model. The goal is to develop algorithms to detect harmful tides in low-performance devices.
3. Solving Process
Change settings for low-performance devices (Lotte Panda)
It has been confirmed that the actual detection speed can be improved without the need to modify or relearn the current model using input dimension changes and multi-core environment of the instrument, as well as the ability to reduce latency by adjusting the number of frames. Currently in preparation for field testing by applying this method directly to the prototype, Using SSD MOBILENET V1 Models
Using SSD MOBILENET, one of the most well-known deep learning techniques related to object detection, we have confirmed that the frame processing speed is higher than that of the existing use model, and we expect to get improved results from the actual low-performance device.
4. Ripple effects and future plans
If excellence is proven in the field, it is expected that economic wealth will be created not only for companies but also for the nation through expansion of regions across the country, expansion of overseas markets through diversification of products, and promotion of projects to link harmful tidal waves.