microsoft / yolat-vectorgraphicsrecognition Goto Github PK

Dr.Jiang,
Sorry to bother you.
While performing the diagrams dataset replication, I found that the 'outlets' are not detectable at all.
And the results are much different from the results obtained in your paper, my results are as follows:
[email protected]: 0.9275 [email protected]: 0.8790 MAP@ALL: 0.8392
While executing build_graph_bbox_diagram.py, I found that the generated bbox divides two 'outlets' in the same bbox, I modified the parameter expand_length=(15 / width, 30 / height), the generated bbox was able to differentiate the two 'outlets' but it was still recognized as none in the training results.
I find this phenomenon very strange and hope to get your answer.
Thanks a lot and looking forward to your response soon.

Best regards
geruiLin

Dr.Jiang,
Sorry to bother you.
I run the command "CUDA_VISIBLE_DEVICES=0 python -u cad_recognition/test.py --data_dir data/FloorPlansGraph5_iter --pretrained_model log/run182_2_best.pth" with codes about "opt.arch" and "opt.graph" being commented out.
BUT before and then, I still got the errors:
"size mismatch for cls_net.fusion_block.0.weight: copying a param with shape torch.Size([1024, 128]) from checkpoint, the shape in current model is torch.Size([1024, 448]).
size mismatch for cls_net.fusion_block_super.0.weight: copying a param with shape torch.Size([1024, 128]) from checkpoint, the shape in current model is torch.Size([1024, 448]).
size mismatch for prediction_cls.0.0.weight: copying a param with shape torch.Size([512, 2304]) from checkpoint, the shape in current model is torch.Size([512, 2944])."
It really confusing since the model was saved based on "def save_checkpoint()" while it did not match during loading the model.
Would you like to resolve this issue?
Thanks a lot and looking forward to your response soon.

Best regards,
VivianBB.

1. thop, h5py and fvcore should be included in some requirements.txt somewhere or in the deepgcn_env_install.sh file as these were required to prepare the Floorplan dataset

2. any line with np.bool in the build_graph_bbox.py file should be changed to simply bool. Error is:
   "AttributeError: module 'numpy' has no attribute 'bool'.
   np.bool was a deprecated alias for the builtin bool. To avoid this error in existing code, use bool by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use np.bool_ here."

With the first two fixes, the data preparation can be done without errors for me while following the introductory steps.

3. In cad_recognition/train.py, the "keys" of the collate function (line 126) should be defined as "data_list[0].keys()" instead of "data_list[0].keys". The original definition assigns a method to the variable "keys" and causes an error when enumerating the train_loader.

With the above fix in step 3, the training ran without a hitch for me while following the introductory steps.

Dr.Jiang,
Sorry to bother you.I am a novice and I am not very familiar with how to use the trained model for new predictions. Could you please explain to me how to do that? I have followed the operation steps on Git to train the model and obtained the file "run182_2_best.pth". So, how can I use this training result to predict new files? Thank you.