Portfolio

Started at 2017 and officially founded in  2019, we have worked on different use-cases in the field of AI for customers. Normally projects start with a fast P.o.C (Proof of Concept) phase. Once the prototype is functioning and the minimum functionality is there, we continue the evolutionary loop of development with a focus on software and process standards. We give our best to deliver codes that are well documented and reusable. Using project management tools such as Jira we make sure the work is on track and delivery will be on time and budget. Machine Learning projects are in nature very uncertain. This is however in contrast to customer demands when it comes to delivering on time and modules with demanded accuracy and performance. To tackle this challenge, we try to stay in a “real-time” contact with  customers so that they have a clear idea of the progress. We also use our experience and know-how from the past to choose approaches which have more success probability. 

Plate_Detector_Anonymizer
Plate Detector and Anonymizer
We developed our license plate detector based on the state-of-the-art technologies. Our license plate detector can be fine-tuned for different license plate types. It is fast and robust which work on both CPU and GPU.
Capabilities
  • Fast and accurate license plate detection
  • Multiple license detection
  • Anonymizing the license in the determined zone
  • Trainable for new license type
  • Work online (Now – about 30 fps on GPU – about 3 fps on CPU)

Applications
  • Video Surveillance
  • Privacy solutions
Innovations & Technologies
  • Deep learning based solution
  • Optimizing the inference time of model
  • .Net framework
MultiFace Recognizer
Multi-Face Recognizer on the Wild Videos
We developed a robust and accurate deep learning-based solution for detection and recognition of the face in the wild videos. Our face detector is based on the state-of-the-art face detection method which work well on the wild, and we develop a more accurate technology for robust face recognition.
Capabilities
  • Implemented robust face detector
  • Using state-of-the-art deep learning based face recognizer
  • Work with input a few images from the target (although one image!)
  • Create embedding ID for each face in database
  • Supporting small faces by using Super-resolution techniques
  • Supporting frontal and side-view faces
  • Work offline (Now - 2 fps in online on GPUs)
Applications
  • Video surveillance
  • Searching a person on the long time videos
  • Face anonymizing for specific persons along IP cameras
  • Face based identifications
Innovations & Technologies
  • Combining several state-of-the-art facial technologies
  • Applying deep learning-based super resolution techniques
  • Developed in the Pytorch framework
  • Based on GPU
  • Docker supporting
OCR_pdf_file_translation
OCR pdf file with translation​
The goal of this project is converting pdf files in different languages to text files in other languages. There were some challenges as number of pdf pages, quality of pdf and etc.
Capabilities
  • Splitting pdf file to its pages
  • Convert pdfs to image format
  • Improving image quality
  • Translating to any language by using free google API
Applications
  • Digitalizing paper documents and saving them in a convenience way.
  • Robust OCR for low quality files
  • Translating forms and documents and sending them to the customers in any country.
Innovations & Technologies
  • Finding an optimized method to convert pdf files to images file since Tesseract operates better.
  • Using preprocessing methods to leverage the image quality.
AI-Bridge OCR Deploy Process
Characteristics of AI-Bridge-Powered OCR Method

Characteristics
  • Powerful Deep Learning based text detector: High, Accurate detector, Robust in the sense of noise
  • Deep learning based optical character recognition
  • Applying several Deep Learning based Optical Character recognition and ensemble them
  • Robust against illumination variant
  • Good Accuracy in the sense of poor scanned file
  • Applying NLP rules for preprocessing of the extracted text
  • Formatted JSON output text
  • English and German language supporting
  • Trainable on the new characters
Applications
  • Extracting information from documents: Cheque, passport, invoice, bank statement and receipt
  • Automatic number plate recognition
  • etc.