kentaroy47
commented
4 years ago 8 and readme has some insights!
You can use voc format or txt files as well. If you still have trouble, please tell uss!
You can either try voc or simple parsers for your dataset.
simple parsers are much easier, while you train your network as:
python train_rpn.py --network vgg16 -o simple -p ./dataset.txt Simply provide a text file, with each line containing:
filepath,x1,y1,x2,y2,class_name For example:
/data/imgs/img_001.jpg,837,346,981,456
The code expects to be used on datasets already there based all in H5 files
We are able to create H5 datasets from pictures we want to train, but how do we include bounding boxes in those images?
In VOC and COCO is quite straightforward (XML and JSON files respectively), but with H5 we are at a loss :(
If someone could shed some light that´d be very appreciated