Closed SherlockHua1995 closed 5 years ago
Hey @SherlockHua1995, here's the link for LabelImg (https://github.com/tzutalin/labelImg) This tool is pretty straightforward to use and will give you the coordinates of bounding boxes as XML files for each image, which can be converted to a csv and then into a TFR format. The csv and TFR conversion code is available in the official Tensorflow repo. You can also have look at these links for the same:
Hope this helps!
Thanks for your reply. Is the train set in your repository made by yourself ? And what size of the training set is sufficient for a accurate model to count the people ? yours~
Hey! Yes the train set is created by me using LabelImg. The size of the training set depends more your use case. You could try will about a 100 annotated images and add images as per the required accuracy. Thanks!
Hello, thanks for your code. I am working on this field too and I want to build a project that counting people under a surveillance camera. How to use a large dataset to train the model? Although you have mentioned in readme , I didn't found the LabelImg . Any help would be appreciated.