Error when training VOC dataset (non-augmented)

[vioanl]

I encountered an error when trying to train the model on the Pascal VOC dataset (non-augmented version) using the command:sh scripts/voc/rcil_10-1-overlap.sh for the voc dataset but not the augmented version. The error suggests that the dataset indices are out of range, and I am unsure how to properly prepare my dataset to avoid these errors:

Sample entries from /home/.../PascalVOC12/list/train.txt:

Entry 0: 2007_000032
Entry 1: 2007_000039
Entry 2: 2007_000063
Entry 3: 2007_000068
Entry 4: 2007_000121
Total images loaded: 1464
Subset: dataset size = 1464, max index requested = 10581
...
ValueError: Invalid index: max requested index (10581) exceeds dataset size

(1464)!

What is the expected format for train.txt? Should it contain just image IDs or full paths?Are there any pre-processing steps that I should do before training?

[zhangchbin]

Hi, @vioanl
Please delete the preprocessed index at data/voc before running. By default, the code will use the preprocessed index files if they exist. You may refer to this line for more details.