use of base_dir

[velastin]

I am not sure I am doing something wrong. I am checking if I can train with the data on a different location. I have everything (images, annotations, ...) in subdirectories of a new directory MyResNetEM. I then execute the following

python -u object_detector_retinanet/keras_retinanet/bin/train.py --gpu 0 --tensorboard-dir MyResNetEM/logs --snapshot-path MyResNetEM/snapshot csv --base_dir MyResNetEM/JPEGImages --annotations MyResNetEM/EMARetNetEM-annotations_train.csv --val-annotations MyResNetEM/EMARetNetEM-annotations_val.csv

I get an error that I traced to order.sort(key=lambda x: self.image_aspect_ratio(x)) in generator.py. image_aspect_ratio() is in csv_generator.py. It calls image_path() also in csv_generator.py

image_path() returns os.path.join(self.base_dir, self.image_names[image_index])

but I found that the elements of image_names[ ] already contain the base_dir, so image_path adds a second unnecessary base_dir resulting in the failure ...

I have modified image_path to simply return the appropriate element of image_names [ ]
The scripts seems to be happy now (training), but I am concerned this could have introduced another problem.

Comments welcome ...