Deep learning for sport video segmentation (repost)

Hello, I believe my skill in this background will prove to be of great help to you. Please contact me to discuss more, Waiting for your kind reply. Thank you

Hello, I have experience in using neural networks for video processing and will be glad to help you with your project. I think it can be done using time series rnn to detect the "ball is in play" But I need more information from you about it: 1. Can you give the example of video 2. Will there be just one static camera or several cams from different point of view? 3. Do you have labeled dataset of videos? 4. If not who will label the data? 5. Not sure about using audio stream, need your explanation Will be glad to hear from you. Best regards, Andrey Kropachev

This is video segmentation which is essentially processing of sequential data and splitting the data in time domain. We need to use recurrent neural network models for processing this data. We propose using this in tensor flow with a keras wrapper First phase we will do offline on a normal cpu then depending upon the extent of data we will shift towards online training Multiple itiration will be given to you ascertain the quality of output

Hi! My name is Konstantin. I'm an expert in machine and deep learning and I have a lot of experience (outside of freelancer) with Tensorflow and Keras running on top of GPU accelerators user for various objects detection, video and image frames classification . The project I'm working on right now is real-time video frames classification using Tensorflow, my algorithm is classifying whole frames using a combination of transfer learning and custom deep-network approaches and another one doing objects recognition on selected frames. I also have a lot of experience with C/C++, OpenCV and different kinds of image recognition and computer vision problems. So I think I can do this task. But before making a final conclusions about the price and time frame, I would like to know and discuss specific details about your needs. And one of particular questions - do you have a dataset with video or frames can be used for a model training or the data should be captured and/or annotated first? If you do have annotated data - how big is it? Next, are you aiming real-time or off-line processing is fine? With best regards, Konstantin