This is a dataset of photos from Aliexpress product reviews. 379186 images, ~30Gb of data. The dataset is great for example for training a model to find similar images.
First, you have to download images, this requires a script download.py.
Some EDA and photos you can found here
| Column | Description |
|---|---|
| id | Unique identificator of Aliexpress product. Each product can have several colors. |
| category | Clothing category, like skirt |
| color | Color of current thing. The value does not always make sense - it is such as the seller filled it. |
| source | Two value possible for each product: 'shop' and 'customer'. 'shop' is usually a good photo. Customer photo closer to life. |
| url | url to image |
Dataset cleaned semi-manual in a semi-manual way. Some images was annotated good/bad (photo of package for example). Then a model was trained that cleared the rest of the dataset.
If you find bad photos, or you want to add something then feel free to use pull request.