This is our solution for Perfect Half Million Beauty Product Image Recognition Challenge , which obtained the 1st place (USTC_NELSLIP) with MAP@7 0.4086

• python3.5
• pytorch(1.1.0)
• torchvision
• pretrainedmodels
• tqdm
• PIL

1.Download all the images of dataset recorded in csvfile

2.Transform all the images to square form using 255 padding

3.Put all the images in the folder data_clean

Query images are put in the folder search/ori_images

SEResnet152 Densenet201

Download these two models and put them in the folder /code/pretrained

1. Extract features of images in database(Perfect-500K)

$ python features.py

2. Conduct image retrieval

$ python search.py

All the models are all pretrained on the Imagenet. They can be downloaded from module pretrainedmodels.