This package is based on MMdetection v2.0.
MMdetection installation and usage guidelines are provided in it's documentation. A slightly modified dockerfile is also provided here.
To run the detector and test on images or videos, two main components are needed. The first one are the configuration files which describe which detector is used. The second component are the trained models using that specific detector. When providing the input to MMdetection, select the proper configuration file with its respective trained model. The list below describes that. Use example to test an image or video as in here.
The path to the trained models is this: https://mirror.vtti.vt.edu/vtti/ctbs/object_detection/
Exact correspondence is as follows:
VA Beach Traffic Dataset
Config file: custom_configs/va_beach/faster_rcnn_x101_32x4d_fpn_1x_coco.py
Model file: va_beach.pth
Classes: classes_traffic.txt
SHRP2 original split
Config file: custom_configs/shrp2/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py
Model file: SHRP2_original_split.pth
Classes: classes_shrp2_train.txt
Extra Data only:
Config file: custom_configs/shrp2+extra/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py
Model file: extra_data_only.pth
Classes: classes_extra.txt
SHRP2 original split + Extra Data (Outside Objects set)
Config file: custom_configs/shrp2+extra/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py
Model file: shrp2_original_split+extra_data_outside_objects.pth
Classes: classes_shrp2_extra.txt
SHRP2 (with more SHRP2 images on test set) + Extra Data (Outside Objects set)
Config file: custom_configs/shrp2+extra/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py
Model file: shrp2_less_on_train_set+extra_data_outside_objects.pth
Classes: classes_shrp2_extra.txt
Extra Data (Billboards only)
Config file: custom_configs/billboards/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py
Model file: billboards_only.pth
Classes: classes_billboard.txt
Extra Data (Inside Objects set), each config file corresponds to each model and classes file with the same order.
Config file:
custom_configs/extra_inside_objects/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco.py
custom_configs/extra_inside_objects/cascade_rcnn_r101_fpn_dconv_c3-c5_1x_coco2.py
custom_configs/extra_inside_objects/htc_x101_32x4d_fpn_16x1_20e_coco.py
Model file:
more categories: inside_objects_1.pth
less categories: inside_objects_2.pth
less categories with mask: inside_objects_mask
Classes:
classes_inside_objects.txt
classes_inside_objects2.txt
classes_inside_objects2.txt