Human-Object Interaction Detection via Global Context and Pairwise-level Fusion Features Integration
Training inference (vscode debugging scripts are provided and can be used directly)
. /experiments/ contains scripts for foreground & background running, and scripts for waiting for the graphics card to idle.
For pre-training models, please refer to the UPT (https://github.com/fredzzhang/upt)
hicodet\hico_20160224_det Here is the HICODET dataset (refer to UPT).
pocket repository I modified a little to support multi-threaded training (https://github.com/fredzzhang/pocket)
vcoco\v-coco put vcoco repository (refer to UPT)
Fix previous issue of labels not counting images without HOI (VCOCO/HICO-DET onedrive) https://1drv.ms/f/s!AmZ_JQSzOHiqlZ8H5FaDuUSZ5Th7bg?e=PMMtQI
logs to put the output model
Location of trained model (VCOCO/HICO-DET onedrive) https://1drv.ms/f/s!AmZ_JQSzOHiqla8FNyxEJIiSqqz62A?e=txzh8q
Pytorch using 1.7.1 + cuda10.1.
If there are any problems you can contact: ddwhzh@163.com