search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.67k stars 11.71k forks source link

A Question about Data #1165

Open zhouhao-learning opened 5 years ago

zhouhao-learning commented 5 years ago

I want to know, if I do image target location detection, then my training data must be labeled one by one? For example, I want to detect the people and cars in the image, so when I train, do I have to label the people and cars in the training data?

stygian2a commented 5 years ago

Yes of course! You might find already labeled data online though

zhouhao-learning commented 5 years ago

@stygian2a Can there be any algorithm to replace this annotation? Because if it's a cell image, it's too hard to label one cell at a time.

stygian2a commented 5 years ago

No, I'm sorry. There are some software that helps you make manual annotations though. Can't you find an already annotated dataset?

Besides, if you had an algorithm that automatically labelled your images, why would be the use of the training? You would already have a program that detect whatever you are trying to detect !

HAMZARaouia commented 5 years ago

The are some tools for manual labeling like Labelme if you need the semantic segmentation otherwise you can use LabelImg to do only detection