ksaito-ut/openworld_ldet

openworld_ldet's Issues

agnostic setting

Hi, I have two simple questions.

1. How to set config file when train plain maskrcnn in an agnostic way.
2. why set num_classes to 80 under VOC to NonVOC setting.

Version issue of Mapillary Dataset

Hi, I have some confusion about the versions of the mapillary dataset.

1. In the paper you claim you have used 35 foreground classes of the mapillary dataset. However, this should correspond to v1.2 according to official instructions.

The Mapillary Vistas Dataset is a large-scale street-level image dataset containing 25,000 high-resolution images annotated into 66/124 object categories of which 37/70 classes are instance-specific labels (v.1.2 and v2.0, respectively).

2. The json file of the mapillary dataset you use seems to come from [repo] (https://github.com/Luodian/Mapillary2COCO), this also corresponds to v1.2.
3. However you seem to use v2.0 in your code.

_PREDEFINED_SPLITS_MAP['map'] = {"map_val": ("path/mapillary-vistas/mapillary-vistas-dataset_public_v2.0/validation/images",

Question about criteria for UVO evaluation videos

First of all, thank you for your research.

I have analyzed the provided UVO_frame_val_exist.json, and it appears that evaluation is being conducted on 742 videos. However, the original UVO_sparse dataset's validation json contains information for 2452 videos. I'm curious about the criteria for evaluating only 742 videos, and if there's something I might have missed.

Thank you.

why mask-rcnn-p not work?

Hi,
I notice that the performance of Mask-RCNN-P is very similar to the vanilla Mask-RCNN in table 1. Is there any reason why the pseudo label method like OWOD didn't work?

AttributeError: type object 'Tensor' has no attribute 'cat'

File "/home/subinyi/anaconda3/envs/ldet/lib/python3.7/site-packages/detectron2/modeling/box_regression.py", line 327, in _dense_box_regression_loss
anchors = type(anchors[0]).cat(anchors).tensor # (R, 4)
AttributeError: type object 'Tensor' has no attribute 'cat'

i dont know how to solve this issue? can you help me

What is the evaluation code bug?

Hi,

I notice that you update the README and said there is an evaluation code bug. Could you plz give more information about this?

Visualization demo?

Hi,

Would you plz give some instructions on how to visualize the LDET results?