Maintaining Users Clothing Wardrobe With YoloV5

In this project, I have created a deep-learning based system that can understand which clothing items is wear by the users and, based on that it makes the decision whether the clothing item is already present in the online wardrobe or not and if not, it will automatically get added into the wardrobe by the system.

Libraries Used

Pytorch
YoloV5
Numpy
Pandas
Matplotlib
Sklearn
Python Imaging Library
OpenCV
skimage

Dataset

For clothing item identification and localization, I have created a custom dataset by myself. The clothing item images have been downloaded from Kaggle's InShopeClothes dataset and it gets annotated with a bounding box with the open-source tool called YoloLabel. A total of 656 images with their respective label in the dataset, out of which 586 images with their label are in the training section and the remaining 66 are in the validation section. The dataset can be downloaded from here: https://bit.ly/3eEt90Y

Object Detection

For object detection, I have used here YoloV5 mainly for its smaller size, faster computation, and state-of-the-art accuracy. The model has to get trained over 586 images for 50 epochs. The model took 21 minutes to train completely, which means it takes 41 seconds per epoch to complete its task. The model was able to perform object detection with the following accuracy.

Object Detection Result On Validation Dataset

Input:

Output:

Image Similarity

For making the system understand whether the images are similar or dissimilar, here I have used Structural Similarity Index (SSIM) for scoring the similarity between the two images. The SSIM attempts to model the perceived change in the structural information of the image and based on that generates a score.

Online Wardrobe Management

For the management of online wardrobe and putting in the newly detected clothing item, I have written a logical code that compares the detected clothing item to preexisted clothing item within the wardrobe and generates an SSIM score, if the score is above a certain threshold the item will get added to the wardrobe, else not.

Current Wardrobe:

User Clothing Item:

User Clothing Item Detected:

Updated Wardrobe:

Conclusion

Here in this project, I have created a system that uses YoloV5 for identifying the clothing item and put them into the online wardrobe if that clothing item is not present.

References

https://www.kaggle.com/rvnrvn1/inshopclothes
https://github.com/developer0hye/Yolo_Label
https://github.com/ultralytics/yolov5
https://medium.com/srm-mic/all-about-structural-similarity-index-ssim-theory-code-in-pytorch-6551b455541e

navinbondade / maintaining-users-clothing-wardrobe-with-yolov5 Goto Github PK