Navras

- improved a model for door detection by 10%
- built multitask model for floorplan segmentation and joint detection with 86% IoU and 90% f1 of joints
- developed a post-processing algorithm for NN output to extract components and their relationships to make it easier to take architectural measurements and make adjustments based on user feedback
- developed algorithms for interacting with floorplan labels including: symbol detection by click, which allowed in 80% of cases to select a symbol with a single click, symbol detection by template based on traditional approaches and metric learning with 92% f1, OCR, finding similar symbols and grouping by text and/or shape
- developed generators of synthetic floorplans and architectural symbols, which resulted in an improvement of 18% to 69% for different models
- organized the floorplan annotation process, new data allowed to increase the IoU by 8%
- developed a testing procedure for the final result of the service to compare it with metrics at intermediate stages

improved a model for door detection by 10%
built multitask model for floorplan segmentation and joint detection with 86% IoU and 90% f1 of joints
developed a post-processing algorithm for NN output to extract components and their relationships to make it easier to take architectural measurements and make adjustments based on user feedback
developed algorithms for interacting with floorplan labels including: symbol detection by click, which allowed in 80% of cases to select a symbol with a single click, symbol detection by template based on traditional approaches and metric learning with 92% f1, OCR, finding similar symbols and grouping by text and/or shape
developed generators of synthetic floorplans and architectural symbols, which resulted in an improvement of 18% to 69% for different models
organized the floorplan annotation process, new data allowed to increase the IoU by 8%
developed a testing procedure for the final result of the service to compare it with metrics at intermediate stages

- developed an algorithm that detects free parking spaces and is resistant to non-compliance with markings
- developed a free parking space verification system based on classic approaches to reduce FPs by 15%
- developed a car monitoring module
- developed data collection/annotation pipeline to create a dataset
- built model for car segmentation with 93% mAP
- developed back-end that served 100 cameras
- Improved the user experience and safety for drivers by developing a skill for Voice Assistant as an interface

- developed an algorithm that detects free parking spaces and is resistant to non-compliance with markings
- developed a free parking space verification system based on classic approaches to reduce FPs by 15%
- developed a car monitoring module
- developed data collection/annotation pipeline to create a dataset
- built model for car segmentation with 93% mAP
- developed back-end that served 100 cameras
- Improved the user experience and safety for drivers by developing a skill for Voice Assistant as an interface

developed an algorithm that detects free parking spaces and is resistant to non-compliance with markings
developed a free parking space verification system based on classic approaches to reduce FPs by 15%
developed a car monitoring module
developed data collection/annotation pipeline to create a dataset
built model for car segmentation with 93% mAP
developed back-end that served 100 cameras
Improved the user experience and safety for drivers by developing a skill for Voice Assistant as an interface