To train a model with the Tensorflow Object Detection API you'll need a TFRecord file. Unfortunately, that binary format is not human-readable so, traditionally, you would convert to this specific CSV format first and then use a custom script to create the TFRecord.
With Roboflow, you don't need to do that because we can export TFRecords directly. But, if you do want to inspect the contents for yourself, you can still use Roboflow to create the CSV. Or, if you already have your data in the Tensorflow CSV format you can use Roboflow to convert it to another annotation format to dip your toe in the waters beyond the Tensorflow ecosystem.
Below, learn the structure of Tensorflow Object Detection CSV.
filename,width,height,class,xmin,ymin,xmax,ymax
000001.jpg,500,375,helmet,111,144,134,174
000001.jpg,500,375,helmet,178,84,230,143
000007.jpg,500,466,helmet,115,139,180,230
000007.jpg,500,466,helmet,174,156,201,219
000007.jpg,500,466,helmet,197,177,231,227
000007.jpg,500,466,helmet,247,124,294,203
000007.jpg,500,466,helmet,280,127,337,208
000007.jpg,500,466,helmet,336,148,387,223
000007.jpg,500,466,helmet,375,152,410,219With Roboflow supervision, an open source Python package with utilities for completing computer vision tasks, you can merge and split detections in Tensorflow Object Detection CSV. Read our dedicated guides to learn how to merge and split Tensorflow Object Detection CSV detections.
Below, see model architectures that require data in the Tensorflow Object Detection CSV format when training a new model.
On each page below, you can find links to our guides that show how to plot predictions from the model, and complete other common tasks like detecting small objects with the model.