How to visualize with result json file? about deepfashion2 HOT 15 CLOSED

@seokhyeonSong Do you want to visualize annotations of DeepFashion2 or visualize predicted results from network?

from deepfashion2.

@geyuying I found how to train and I'm training it.

what I want to know is how to visualize with predicted result from network.

If I make a pre-trained model, then what should I have to do with which py file?

sorry for just my ignorance...

from deepfashion2.

Our framework is based on Detectron, and we use Detectron/tools/infer_simple.py for visualization. You can refer to https://github.com/facebookresearch/Detectron/blob/master/GETTING_STARTED.md for details.

from deepfashion2.

@geyuying

Thanks for your help

I'll try it!

from deepfashion2.

@geyuying

Sorry for asking again..

I've tried sample_infer.py as python infer_simple.py --cfg ../configs/12_2017_baselines/e2e_faster_rcnn_R-50-FPN_1x.yaml --output-dir ./ --image-ext jpg --wts ../../match_rcnn/logs/deepfashion220190808T0916/mask_rcnn_deepfashion2_0005.h5 ./

but Detectron it seems to be cannot load h5 weight file..

I've got new h5 file by running match_rcnn's main.py

Is h5 file have to be transformed?

from deepfashion2.

from deepfashion2.

@jia0511

run evaluation/deepfasion2_to_coco.py to get json file

you may run this code with validation dataset and train dataset

Then you may get 2 json files.

Then run match_rcnn's main.py

You have to change settings with your environment

Then training will be started

That's what I've got known...

It could be wrong

from deepfashion2.

from deepfashion2.

from deepfashion2.

@jia0511

cuz This git's owner shares datasets to only the people who send google form

and trained model can be obtained by dataset so it will be impossible sry

from deepfashion2.

@jia0511
you have to change

json_name ~~, image_name ~~ (row 108,109) <- the link to dataset's annos folder and image folder

json_name (row 234) <- the name of json file to create

from deepfashion2.

from deepfashion2.

@jia0511

You also change num_images to the number of images train & validation has

from deepfashion2.

from deepfashion2.

I found how to train with sample_infer.py

make json file with deepfashion dataset with mtcnn's main.py

and run with this code https://github.com/facebookresearch/Detectron/blob/master/GETTING_STARTED.md#training-a-model-with-detectron

but dataset have to be changed

change Detectron/detectron/datasets/dataset_catalog.py
-> add new available datasets with deepfashion train & test with image files and json files

change Detectron/configs/getting_started/config_which_you_want
-> change DATASETS with new available datasets' name which you made at dataset_catalog.py

and then run train code with https://github.com/facebookresearch/Detectron/blob/master/GETTING_STARTED.md#training-a-model-with-detectron then train will be started

I've just started train but it would be make it visualize with the weights file that made with this code and config files that provided with detectron

Thanks for great help @geyuying

from deepfashion2.