ihciah / deep-fashion-retrieval Goto Github PK

I use a picture contains in my dataset to test. And found that Kmeans has a wrong cluster that don't contain my test picture feature. so I print the features and found the feature generate by the dump_single_feature method is different to the feature generate by the feature_extractor.py.Therefor it can't found the correct picture.Is there any solutions to slove this problem?

When my accuracy become stable,I set FREEZE to False ,but in a minute,the accuracy become stable again。What should I do？

I couldn't open model after downloading from drive.google.
I tried to repair it but fail
How can I do next?

Hi Sir,
I was looking forward to your project and I wanted to use this for my current project. Since I am working for a different organization, so I need the trained model to be able to search from a different database. Also I want to add some more classes in the training model. Is it possible sir?
Thanks a lot.

[Saved]
Hi, ihciah , thank you for your sharing. But I could not download the model from google drive, could you please put one copy of the model to baidu drive?
Thanks!

感谢您的分享！ 我目前也在学校里学习这方面的知识，但通过DeepFashion官方百度云链接下载的数据集需要解压密码....我发送邮件询问原作者，可能他们太忙了两个月过去了也没等到他的回复。所以，可否请您告知一下我解压密码呢？我绝对不会违背原作者意愿私自泄露，仅仅是自己学习之用。以防公开您可以单独发送至我的邮箱[email protected] 期待您的回复谢谢！

刚提问完几个小时我就收到了回复，现在已经获取到解压密码了哈哈，不好意思打扰您啦

Can you please explain how to train this on complete dataset? Your model is trained only on 20 categories. As per the instructions in the readme, I have to set freeze to true and train. Then, after 10 epochs. If I change freeze to False, then how to again train?
When I try to train the same model, the accuracy moves from 59% to 3%?
Can you please tell me, how to load my model for retrainig?
Thanks

@ihciah hello, I find, although the batch_size of triplet in config.py is 32, but only one triplet is loaded during training, hence, the tripleloss often get zeros. Larger batch size of triplet hurts the classification? or other reasons? Thank you very much.

你好 我在运行代码过程中出现了问题 想请教一下 方便加v 15591987380 谢谢
看你是选取了20类(上衣），那么所有Anno中标签也要改动吗？
我在运行中出现 BrokenPipeError: [Errno 32] Broken pipe 是怎么回事呢

Hello, thanks for devoloping your the deep-fashion-retrieval project. I need a little help for running your project.

I have following files:

DATASET_BASE
+Anno
+Eval
+img
+models: (only your model_10_final.pth.tar)
+in_shop(optional)
+list_bbox_inshop.txt
+list_eval_partition.txt
+img
+all_feat.list(generated)
+all_feat.npy(generated)
+all_color_feat.npy(generated)

When I run the "python retrieval.py img/Sheer_Pleated-Front_Blouse/img_00000005.jpg", it is occuring following errors:

/usr/local/lib/python3.6/dist-packages/torchvision/transforms/transforms.py:188:
UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead. "please use transforms.Resize instead.")
Loading model...
Loading model Done. Time: 4.231 sec
Loading feature database...
Loading feature database Done. Time: 0.738 sec
Extracting image feature...
Extracting image feature Done. Time: 0.480 sec
Loading feature K-means model...
Traceback (most recent call last): File "retrieval.py", line 147, in
clf = load_kmeans_model() File "/content/drive/My Drive/Deep_Learning/deep-fashion-retrieval/utils.py", line 110, in wrapped
ret = fn(*args, **kw) File "retrieval.py", line 45, in load_kmeans_model
clf = joblib.load(clf_model_path) File "/usr/local/lib/python3.6/dist-packages/sklearn/externals/joblib/numpy_pickle.py", line 590, in load
with open(filename, 'rb') as f: FileNotFoundError: [Errno 2] No such file or directory: 'DATASET_BASE/deepfashion_data/models/kmeans.m'

When I run the kmeans.py, it is occuring following errors:

/usr/local/lib/python3.6/dist-packages/torchvision/transforms/transforms.py:188:
UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead. "please use transforms.Resize instead.")
Loading feature database...
Loading feature database Done. Time: 0.889 sec
Traceback (most recent call last):
File "kmeans.py", line 12, in
feats, labels = load_feat_db() ValueError: too many values to unpack (expected 2)

How to create kmeans.m? Or how to predict my special image? Thanks.

Hello,

I would like to know whether you just use the classification information (1, 2, 3 for full, upper, lower) for the loss function, instead of the bbox and other information?

It looks like you freeze all the conv layer parameters?

您好，dataset百度云盘链接现在是无效的，可以重新分享下链接吗？

Hi@ihciah
can you please tell me your deep fashion model is applicable for continuous learning and transfer learning for inclusion of new categories.

Best,
Himabindu Y

First of all, thank you very much for sharing your code! I have a question for you.
When I ran kmeans.py, the system prompted the error: too many values to unpack. If I use the "Deep Feature Weighted-Color Feature" method, what two variables should the return value of the "load_feat_db" function be, and what do I need to change "return deep_feats,color_feats,labels" to?

Hey,

great work you did there. Could you specify the license under which this code is available?