What is ODHUB?
ODHUB is an object detection studio that allows users to design, train, and validate object detection models without coding.
With ODHUB, you can create smart computer vision tasks, such as defect detection in manufacturing, license place recognition/detection, people counting, food detection, etc.
ODHUB supports the TensorFlow framework (SSD, Faster RCNN, EfficientDet) and YOLO framework (YOLO2, YOLO3, YOLO4) for the highest inference speed and accuracy.
Which dataset format does ODHUB support for training?
A sufficient training dataset to load into ODHUB includes:
- Image files (jpg, png, or bmp format); recommended 300 x 300-pixel image at minimum
- Label file for each image (PASCAL VOC format)
- Label map file to specify label ID and category (.pbtxt format)
What's the difference between ODHUB and DLHUB?
ODHUB focuses on computer vision tasks that detect and localize objects on video streams or images.
DLHUB is more about generic deep-learning applications. You can use DLHUB for regression (prediction) and classification for any data, including images (without giving you a bounding box), audio signals, or tabula data formats.
In terms of detection accuracy, DLHUB and ODHUB are similar.
In terms of inference speed, DLHUB might be much faster than ODHUB, as you only need to deal with classification problems (not localization problems).