rantaimu / m-ce2p Goto Github PK

Hello,do you have a result that Box Map about Maskrcnn on LV-MHP-v2 val ?

i want to ask when i try to test the results on the CHIP datasets, why the image_dir path is contained the image for VOC datasets in your M-CE2P-test.py. And when i download the weight for inference,the list file should choose the testlist? the list point to which dataset? LIP or CHIP?

Hi~ @RanTaimu Thank you sharing models and codes. Would you please offer a Google Drive download link? Baidu Drive is not available in my place.
P.S. I tried to retrain the CE2P model on CIHP dataset with randomly initialized layer6 and layer7. But after 20 epochs, the loss still failed to converge. Have you met this problem before? Thanks!

I have downloaded the pretrained model and the related bbox/mask provided by your. However, I get the bad results on LV-MHP-v2 and CIHP val dataset. @RanTaimu @liutinglt

HI, Thanks for your great work.

I wonder if you can share your mask-rcnn configuration like backbone or even the trained weights?

Thank you in advance!

Thanks for your great work!
I'd like to rerun this work first, but I can't install caffe2 for detectron (have tried many tutorials ...), could you help me? Thanks a lot! @RanTaimu
environment:
cuda9.0
cudnn7
python3.6.9
pytorch0.3.1

Hello.Can you tell me the result on the CIHP val set. The paper BraidNetsay your model mIOU 59.50 and APm 42.83. But I test your model mIOU 60.74 and APm 44.48 without flip and mIOU 60.94 and APm 44.76 with flip.I want to check wether I have some error when testing.
Thanks.

Hi, I want to train mask-rcnn on CIHP, but find no training process in MHP_infer_simple_train.py (e.g. loss computing or backward).
Could you please help me?

Dear RanTaimu,
I can not run this file on GPU. When running, only CPU core was used with an input. Ram (16GB) was full. I also change global_model.cuda() to global_model.cuda(0), gt_model.cuda() to gt_model.cuda(0), mrcnn_model.cuda() to mrcnn_model.cuda(0). and then set dataloader with num_workers = 4.
But when i saw monitor, only cpu core was used and my gpu was not used.

I run on MHP v.2
Help me!

Hey friends,
I can not train or evaluate M-CE2P.
For evaluating, i also downloaded weights. But i think the current source doesnt support for MHPv2.0 (for only LIP CE2P). I also change dataset to dataset_for_mhp. Many information is lose as label

Can you show me or point me how to train/evaluate your source?
Thank you so much!