Now let’s see how I went from having many Pikachu images to a nice and tidy dataset readable by TensorFlow. This can be done by executing: # From tensorflow/models/research/ export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slimįor more details about all the dependencies needed, please see the official documentation, available here: Lastly, you need to add the libraries to the PYTHONPATH. If you do not have Protobuf installed, you can download it from here: Once you have cloned the repo, navigate to the “ research” directory and execute: # From tensorflow/models/research/ protoc object_detection/protos/*.proto -python_out=. Next, clone the repo where the Object Detection API is included. Otherwise, see the instructions here on how to install it. Now that we know a bit about the system used for this experiment, I will proceed to explain how you can build your own custom model.īuilding your own custom model Installationīefore we begin, please make sure you have TensorFlow installed on your computer. Moreover, the library also provides several already trained models ready to be used for detection, the option to train in Google Cloud, plus the support of TensorBoard to monitor the training. The library includes many out-of-the-boxes object detection architectures such as SSD (Single Shot Detector), Faster R-CNN (Faster Region-based Convolutional Neural Network), and R-FCN (Region-based Fully Convolutional Networks), as well as several feature extractors like MobileNet, Inception, Resnet these extractors are really significant because they play a huge part in the speed/performance trade-off of the system. According to the documentation and the paper that introduces the library, what makes it unique is that it is able to trade accuracy for speed and memory usage (also vice-versa) so you can adapt the model to suit your needs and your platform of choice, such as a phone. This package is TensorFlow’s response to the object detection problem - that is, the process of detecting real-world objects (or Pikachus) in a frame. Yep, that’s a Pikachu (screenshot of the detection made on the app) Tensorflow Object Detection API
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |