xushilin1 / fashionformer Goto Github PK

Hello folks, that’s a really nice project! I would like to use it in one of my projects, but I saw that this repository don’t have any license file. Do you folks intend to add some open source licensing in this project, or should we assume that the code published here can’t be used in other places?

Thank you 😄

How does the mapping between fashion class and fashion attributes work?
I see the bounding boxes belonging to the fashion class prediction.
And I see 100 attribute arrays.
However, the attributes do not correspond to the fashion classes when mapped over the bounding boxes.

Hi!

Is is possible to upload model weights to some other cloud storage (e.g. gdrive)?

Great work!

The raw model output structure is not clear.

Let's say I run this code:

from mmdet.apis import (async_inference_detector, inference_detector,
                        init_detector, show_result_pyplot)

config="configs/fashionformer/fashionpedia/fashionformer_swin_b_mlvl_feat_6x.py"
checkpoint="fashionformer_swin_b_3x.pth"

# build the model from a config file and a checkpoint file
model = init_detector(config, checkpoint, device="cuda")
# test a single image
result = inference_detector(model, "test.jpg")

I am trying to plot segmentation, classes, and attributes myself. However the result is does not make sense to me.

result is a list of length 3, why?

In : len(result)                                                                                                                                                   
Out: 3

All 3 of them are of length 46, I assume because there are 46 classes?

In : for res in result: 
...:     print(len(res))                                                                                                                                                          
46
46
46

Now looking at each element of the result.
result[0] is a list of Nx5 matrices? Why 5?

    In : for r in result[0]: 
    ...:     print(r.shape) 
    ...:                                                                                                                                                               
(3, 5)
(9, 5)
(2, 5)
...
(0, 5)
(0, 5)
(0, 5)

result[1] and result[2] are the lists of the same lengths:

    In : for r in result[1]: # or result[1]
    ...:     print(len(r)) 
    ...:      
    ...:                                                                                                                                                               
3
9
2
...
0
0

Each sub element of result[1] is (1500, 1500) matrix, while each sub element of result[2] is vector if (294,)
What are these?

hello,I run the demo code.But all the coordinate of BBOX is zero,like this:
[array([[0. , 0. , 0. , 0. , 0.79686284]],
dtype=float32),
array([[0. , 0. , 0. , 0. , 0.05989303]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.0524485]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.02910043],
[0. , 0. , 0. , 0. , 0.02114534],
[0. , 0. , 0. , 0. , 0.01676908],
[0. , 0. , 0. , 0. , 0.01669254]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.02301841],
[0. , 0. , 0. , 0. , 0.01924651]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.03298811]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.01731287]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.89752316]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.06933805],
[0. , 0. , 0. , 0. , 0.05986979],
[0. , 0. , 0. , 0. , 0.02281323],
[0. , 0. , 0. , 0. , 0.01794562]],
dtype=float32),
array([[0. , 0. , 0. , 0. , 0.90336406],
[0. , 0. , 0. , 0. , 0.5355902 ],
[0. , 0. , 0. , 0. , 0.33188647],
[0. , 0. , 0. , 0. , 0.03402795],
[0. , 0. , 0. , 0. , 0.02057427],
[0. , 0. , 0. , 0. , 0.01929876]],
dtype=float32),
array([[0. , 0. , 0. , 0. , 0.77538824],
[0. , 0. , 0. , 0. , 0.7699898 ],
[0. , 0. , 0. , 0. , 0.04109826],
[0. , 0. , 0. , 0. , 0.03557413],
[0. , 0. , 0. , 0. , 0.03185431],
[0. , 0. , 0. , 0. , 0.03024532],
[0. , 0. , 0. , 0. , 0.02898962],
[0. , 0. , 0. , 0. , 0.02549006],
[0. , 0. , 0. , 0. , 0.02261182],
[0. , 0. , 0. , 0. , 0.02258267],
[0. , 0. , 0. , 0. , 0.02063485],
[0. , 0. , 0. , 0. , 0.02014134],
[0. , 0. , 0. , 0. , 0.01998736],
[0. , 0. , 0. , 0. , 0.01909333],
[0. , 0. , 0. , 0. , 0.01787589],
[0. , 0. , 0. , 0. , 0.01767974],
[0. , 0. , 0. , 0. , 0.01759748],
[0. , 0. , 0. , 0. , 0.01683186]],
dtype=float32),
array([[0. , 0. , 0. , 0. , 0.02470406],
[0. , 0. , 0. , 0. , 0.0189336 ],
[0. , 0. , 0. , 0. , 0.01676438]],
dtype=float32),
array([], shape=(0, 5), dtype=float32),
array([[0. , 0. , 0. , 0. , 0.03116137]],
dtype=float32),

Hi, thanks for sharing your code! I spent more than 3 days to train the Fashionformer with swinb backbone. The training config is configs/fashionformer/fashionpedia/fashionformer_swin_b_mlvl_feat_6x.py and I use 8 V100 GPUs to train fashionpedia 12 epochs. Is the training time normal? And I wonder to know how long when you train the above configuration with 3x schedule? Thanks a lot!

Hello, Im a student who studies fashion ai. Thank you for posting the code and cool model.

Hi, sorry to bother you. Your work is really magnificiant. But here I get some questions about the attribute recognition part. In this project, is there any code that can output the attribute recognition of each image? Or if there isn't, could you please teach me how to get the output of attribute recognition? I'm stuck in achieving the cloth attribute recognition, running out of time,and your project is the only channce for me now. So this is really important to me, and if this is solved, I can pay you back with money if you want. Thank you!!!

Could you please give a reason for the choice of models for comparison experiment as they are not the best model in the area of semantic segmentation?

Hi,

I find ModaNet hard to download. I would like to know whether there is some easy way to get the ModaNet datasets. Or could you provide a zip file of ModaNet?

Best regards,

TANG, shixiang

Thanks for your excellent work. Could you please provide some instructions on how to infer on customize inputs?

I used your trained model with config file ‘fashionformer_r50_modanet.py’ to demo my own img, but face the error below. How could I fix the issus?

(open-mmlab) /mnt/workspace/lyc/FashionFormer-main> PYTHONPATH='.' python demo/image_demo.py /mnt/workspace/lyc/mmdet/demo/demo1.jpg /mnt/workspace/lyc/FashionFormer-main/configs/fashionformer/modanet/fashionformer_r50_modanet.py /mnt/workspace/lyc/FashionFormer-main/fashionformer_r50_3x.pth /home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmcv/utils/misc.py:77: UserWarning: projects.KFashion failed to import and is ignored. UserWarning) /home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmcv/utils/misc.py:77: UserWarning: datasets failed to import and is ignored. UserWarning) Traceback (most recent call last): File "demo/image_demo.py", line 68, in <module> main(args) File "demo/image_demo.py", line 34, in main model = init_detector(args.config, args.checkpoint, device=args.device) File "/home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmdet/apis/inference.py", line 44, in init_detector model = build_detector(config.model, test_cfg=config.get('test_cfg')) File "/home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmdet/models/builder.py", line 59, in build_detector cfg, default_args=dict(train_cfg=train_cfg, test_cfg=test_cfg)) File "/home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmcv/utils/registry.py", line 237, in build return self.build_func(*args, **kwargs, registry=self) File "/home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmcv/cnn/builder.py", line 27, in build_model_from_cfg return build_from_cfg(cfg, registry, default_args) File "/home/pai/envs/open-mmlab/lib/python3.7/site-packages/mmcv/utils/registry.py", line 62, in build_from_cfg f'{obj_type} is not in the {registry.name} registry') KeyError: 'FashionFormer is not in the models registry'

I installed all packages like you wrote but can' work

I would like to know if the code and and the model weights will be public anytime soon.

Hello can you tell model checkpoints right configuration

Im trying like this but with custom image I got empty image
model = init_detector(
config='configs/fashionformer/modanet/mask_rcnn_swin_b_modanet.py',
checkpoint='/data/GarmentQC-dataset/fashionformer_r50_3x.pth',device='cpu'
)

mmdet==2.18.0
mmcv==1.3.18
mmcv-full==1.3.8 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html
python 3.7

Also is the mmdet==2.18.0 correct version because looking from demo/image_demo.py in function show_result_pyplot() param out_file and palette is only in model version starting from 2.24.0, just thinking how to replicate your test if im installing mmdet==2.18.0 then I got error TypeError: show_result_pyplot() got an unexpected keyword argument 'palette'
if im updating mmdet to 2.24.0 I got different errors ;)

Thanks for your great work, is there a colab or hugging face version available for us to test our images? ? I'm really into your work but I don't have any PC/GPU server that can run this code.

I'm trying to separate the clothes from the body, so I want to get the split mask (without overlaying the original image) and the corresponding labels on my own data. I tried to modify the relevant function in image_demo.py, but found it failed.

Dear author,thanks for your contribute for this work and it really inspires me a lot. To extend this work, I want to first train this model on deepfashion and fashionpedia datasets to see its efficiency, but I dont't know how to do. Could you release more instruction on this? Thanks a lot

hello when i run demo code there was some error

python demo/image_demo.py ../dataset/test/03fc99c4deaf14724ef6277dec16b8e8.jpg work_dirs/attribute_mask_rcnn_swin_b_3x/attribute_mask_rcnn_swin_b_3x.py work_dirs/attribute_mask_rcnn_swin_b_3x/latest.pth

how can i solve this error?

Traceback (most recent call last):
File "/home/kshan/anaconda3/envs/fashionformer/lib/python3.8/site-packages/mmcv/utils/misc.py", line 73, in import_modules_from_strings
imported_tmp = import_module(imp)
File "/home/kshan/anaconda3/envs/fashionformer/lib/python3.8/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "", line 1014, in _gcd_import
File "", line 991, in _find_and_load
File "", line 973, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'datasets'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "demo/image_demo.py", line 69, in
main(args)
File "demo/image_demo.py", line 35, in main
model = init_detector(args.config, args.checkpoint, device=args.device)
File "/home/kshan/taehyung/mmdetection/mmdet/apis/inference.py", line 33, in init_detector
config = mmcv.Config.fromfile(config)
File "/home/kshan/anaconda3/envs/fashionformer/lib/python3.8/site-packages/mmcv/utils/config.py", line 343, in fromfile
import_modules_from_strings(**cfg_dict['custom_imports'])
File "/home/kshan/anaconda3/envs/fashionformer/lib/python3.8/site-packages/mmcv/utils/misc.py", line 80, in import_modules_from_strings
raise ImportError
ImportError