Detection of household objects using TensorFlow Object Detection API скачать в хорошем качестве

Detection of household objects using TensorFlow Object Detection API

For this video, we trained a convolutional neural network (a MobileNet v1 from the TensorFlow Object Detection API, pretrained on the MSCOCO dataset) to detect a selection of household objects. Our dataset contained 179 images with a total of 1029 annotated objects separated in 13 classes, with roughly 75 examples for each class. Images were taken with three different cellphones and fed to the network in their native resolutions. The dataset has been made freely available online (link at the end of the description). Training took 8 hours on an NVIDIA GTX 1070. This video was not processed in real-time: it took about ten minutes to generate this four minute video. Object classes used in this video: A can of beer (heineken) A box of tea bags (tea_box) A can of energy drink (monster) Three bottles of juice of different colors (yellow_juice, red_juice, purple_juice) A small bottle of water with Iron Man decorations (iron_man) A box of medicine (medicine) A box of milk (milk_box) A milk bottle (milk_bottle) A box of chocolate milk (chocolate milk) A bottle of shampoo (shampoo) A box of oatmeal (which we labelled as cereal, go figure) The whole process was done in a week by members of the RoboFEI@Home team: https://fei.edu.br/robofei/ Link to the image dataset: https://ieee-dataport.org/open-access... Link to all projects and software related to object detection done by me (including the ones used in this video): https://douglasrizzo.com.br/projects#... Small tutorial of how to train a network, with more useful links: https://gist.github.com/douglasrizzo/...