Where can I find "train_vgoi6_clipped.yaml" and "test_vgoi6_clipped.yaml"?
zdxdsw opened this issue · 7 comments
Hello! Thank you so much for providing the well-structured code!
I'm trying to extract image regions features by python tools/test_sg_net.py --config-file sgg_configs/vgattr/vinvl_x152c4.yaml TEST.IMS_PER_BATCH 2 MODEL.WEIGHT models/vinvl/vinvl_vg_x152c4.pth MODEL.ROI_HEADS.NMS_FILTER 1 MODEL.ROI_HEADS.SCORE_THRESH 0.2 DATA_DIR "../maskrcnn-benchmark-1/datasets1" TEST.IGNORE_BOX_REGRESSION True MODEL.ATTRIBUTE_ON True, but I can't find the download links for two config files specified in vinvl_x152c4.yaml, namely TRAIN: ("visualgenome/train_vgoi6_clipped.yaml",) and TEST: ("visualgenome/test_vgoi6_clipped.yaml",)
What I could find were these two TRAIN: ("visualgenome/train_danfeiX_relation_nm.yaml",), TEST: ("visualgenome/test_danfeiX_relation.yaml",) after downloading visualgenome by path/to/azcopy copy 'https://penzhanwu2.blob.core.windows.net/sgg/sgg_benchmark/datasets/visualgenome' <target folder> --recursive. However, with these two config files, my program would complain KeyError: 'box_features', which seemed to be caused by the returned prediction not including the 'box_features' field.
Any suggestions?
Thanks a lot!
I see. I have to set TEST.OUTPUT_FEATURE to True in the yaml file.
@zdxdsw I am getting same error KeyError: 'box_features', even when I set TEST.OUTPUT_FEATURE to True in the yaml file and also pass TEST.OUTPUT_FEATURE True as a command line parameter. Below is the command that I am running.
python tools/test_sg_net.py --config-file sgg_configs/vgattr/vinvl_x152c4_edited.yaml TEST.IMS_PER_BATCH 2 MODEL.WEIGHT pretrained_model/vinvl_vg_x152c4.pth MODEL.ROI_HEADS.NMS_FILTER 1 MODEL.ROI_HEADS.SCORE_THRESH 0.2 DATA_DIR datasets TEST.IGNORE_BOX_REGRESSION True MODEL.ATTRIBUTE_ON True TEST.OUTPUT_FEATURE True
Did you have to just set any other parameter to resolve the error? I have checked the value of output variable here - https://github.com/microsoft/scene_graph_benchmark/blob/main/maskrcnn_benchmark/engine/inference.py#L38 when passing a batch of images as input. Seems output has only BoxList object and there are no feature outputs.
I have explained about my issue in more details here - microsoft/scene_graph_benchmark#24 (comment).
Any debugging pointers will be really helpful.
@zdxdsw I am getting same error
KeyError: 'box_features', even when I set TEST.OUTPUT_FEATURE to True in the yaml file and also passTEST.OUTPUT_FEATURE Trueas a command line parameter. Below is the command that I am running.python tools/test_sg_net.py --config-file sgg_configs/vgattr/vinvl_x152c4_edited.yaml TEST.IMS_PER_BATCH 2 MODEL.WEIGHT pretrained_model/vinvl_vg_x152c4.pth MODEL.ROI_HEADS.NMS_FILTER 1 MODEL.ROI_HEADS.SCORE_THRESH 0.2 DATA_DIR datasets TEST.IGNORE_BOX_REGRESSION True MODEL.ATTRIBUTE_ON True TEST.OUTPUT_FEATURE TrueDid you have to just set any other parameter to resolve the error? I have checked the value of
outputvariable here - https://github.com/microsoft/scene_graph_benchmark/blob/main/maskrcnn_benchmark/engine/inference.py#L38 when passing a batch of images as input. Seems output has only BoxList object and there are no feature outputs.I have explained about my issue in more details here - microsoft/scene_graph_benchmark#24 (comment).
Any debugging pointers will be really helpful.
Also having the same issue... Can someone please help out?
@zdxdsw I could actually resolve the issue. It was something from my side. I feel setting TEST.OUTPUT_FEATURE to True should be work.
Having followed the steps above, I'm still getting the 'box-features' error even with TEST.OUTPUT_FEATURE=True.
Here's a potential solution for those struggling with the above issue still:
Hi, I wonder how can you set TEST.OUTPUT_FEATURE to True in the yaml file? It seems sgg_configs/vgattr/vinvl_x152c4.yaml https://github.com/microsoft/scene_graph_benchmark/blob/main/sgg_configs/vgattr/vinvl_x152c4.yaml doesn't have any argument OUTPUT_FEATURE.