Hi there, I have a question regarding cross-category evaluation. According to the code, the dataset filter out non-VOC annotations, but this does not filter out images. If the images contain other VOC classes, will they be included in the training set.

Thank you for the amazing work, however, when I train the model using default configs, the validation results are extremely low. Could you provide the training loss curve for comparison? Thanks in advance.

Following the installation steps, while running pip install mmcv-full==1.2.7 leads to this error:

Collecting mmcv-full==1.2.7
  Using cached mmcv-full-1.2.7.tar.gz (231 kB)
    ERROR: Command errored out with exit status 1:
     command: 'D:\Software\Anaconda\python.exe' -c 'import sys, setuptools, tokenize; sys.argv[0] = '"'"'C:\\Users\\intel\\AppData\\Local\\Temp\\pip-install-rmjilsef\\mmcv-full\\setup.py'"'"'; __file__='"'"'C:\\Users\\intel\\AppData\\Local\\Temp\\pip-install-rmjilsef\\mmcv-full\\setup.py'"'"';f=getattr(tokenize, '"'"'open'"'"', open)(__file__);code=f.read().replace('"'"'\r\n'"'"', '"'"'\n'"'"');f.close();exec(compile(code, __file__, '"'"'exec'"'"'))' egg_info --egg-base 'C:\Users\intel\AppData\Local\Temp\pip-install-rmjilsef\mmcv-full\pip-egg-info'
         cwd: C:\Users\intel\AppData\Local\Temp\pip-install-rmjilsef\mmcv-full\
    Complete output (5 lines):
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "C:\Users\intel\AppData\Local\Temp\pip-install-rmjilsef\mmcv-full\setup.py", line 121, in <module>
        major, minor, *rest = cv2.__version__.split('.')
    AttributeError: module 'cv2' has no attribute '__version__'
    ----------------------------------------
ERROR: Command errored out with exit status 1: python setup.py egg_info Check the logs for full command output.

This seems to be avoided by running the commands in the new order:

conda create -n oln python=3.7 -y
conda activate oln
conda install pytorch=1.7.0 torchvision cudatoolkit=11.0 -c pytorch -y
pip install -r requirements.txt
pip install mmcv-full==1.2.7
pip install -v -e . 

As pip install -r requirements.txt seems to satsify the error AttributeError: module 'cv2' has no attribute '__version__' encountered in pip install mmcv-full==1.2.7

but the code of OlnRPNHead shows that the iou sample is not used, as the function get_target return to the objectness_targets_list and objectness_weights_list without sample

Hello, I have a few questions to ask.

1. How are the effects shown in Figure 5 and Figure 6 generated in the paper?
2. If you want to generalize the trained model to other data sets, how should you modify the configuration file or code?

Thank you for your reply.

Could you provide the config of OLN-RPN? I remove the part of rcnn in "configs/oln_box/oln_box.py", but the results are worse than the results in the paper.

Hi, thanks for the inspiring work. When I retrained the model with the config /oln_box, i received this: "AssertionError: The num_classes (1) in Shared2FCBBoxScoreHead of MMDataParallel does not matches the length of CLASSES 80) in CocoSplitDataset " So, should I set the num_classes = 80 instead of 1 when training with COCO? Or there is something else I should do. Thanks!