How to use ReNom IMG ¶
This tutorial will show you how to develop and evaluate model.
How to develop model ¶
This is the homepage of ReNom IMG. In order to develop the model, click the “+Add New Model”.
Parameter tuning ¶
We move to parameter tuning pop up. Default setting is recommended parameter of YOLO(You Look Only Once). We can see the detail in YOLO tutorial at ReNom.jp.We can select algorithm in “CNN Architecture”. In Beta version, we can select only “YOLO”. YOLO regards image as grid cell, so we set the value for “Horizontal Cell” and “Vertical Cell” to original images. In “Bounding Box”, we set how many bounding box per cell is appeared when prediction. We set pixel size in “Image Width” and “Image Height”. When it’s 448, it means the original images are set for 448×448 pixels. YOLO extract a batch of data from dataset for training, which is a “epoch”. Therefore, “Total Epoch” means the number of training. “Batch Size” means how many data is extracted per epoch and “Seed” is a default value for random digit. We set recommended parameter for YOLO as default.
After setting parameter, we click “RUN” to start training.
Each parameters are defined as below;
Horizontal Cell： How many horizontal cell is set to divide original image
Vertical Cell： How many vertical cell is set to divide original image
Bounding Box： How many bounding box per cell is predicted
Image Width： How many pixel is set for the width of original image
Image Height： How may pixel is set for the height of original image
Total Epoch： How many times to implement training
Batch Size： How many data per epoch is extracted
Seed： The default value for random digit