The CHALLENGE: A typical solution is to create machine learning models extracting “feature” vectors or “embeddings” for each item. A k-nearest neighbor search is then performed to produce rankings that list as many of the possible exact matches highest on the list of matches. Technically, your choices for the product representation learning approach form the core of this challenge.
Large-scale product recognition is one of the major applications of computer vision and machine learning in the e-commerce domain. Since the number of products is typically much larger than the number of categories of products, image-based product recognition is often cast as a visual search rather than a classification problem. It is also one of the instances of super fine-grained recognition, where there are many products with slight or subtle visual differences. It has always been a challenge to create a benchmark dataset for training and evaluation on various visual search solutions in a real-world setting.
This motivated creation of eProduct, a dataset consisting of 2.5 million product images towards accelerating development in the areas of self-supervised learning, weakly-supervised learning, and multimodal learning, for fine-grained recognition.
eProduct consists of a training dataset and an evaluation dataset. The training set contains 1.3M+ listing images, with titles and hierarchical category labels. The evaluation set includes 5K query and 1.1M+ index images for visual search evaluation. The high-level structure of eProduct is shown in the following figure.
You are asked to develop models and algorithms to retrieve the “same products” (see below) from a million-scale product database given a “query” image. The ideal solution will retrieve the “same products” as the top items in your rankings. Results will be evaluated by the maco-average “recall@10” score based on our ground truth benchmark data. More details about the metric definition can be found under the “Evaluation” tab.
PRIZES: The winning entrants, or winning teams, if applicable, will be offered the following. Any prizes awarded to a team will be split evenly among team members.
- We offer a total $2,000 USD award split between the top 3 winning teams. The prize will be distributed as follows:
- 1st place takes $1,000
- 2nd takes $600
- 3rd takes $400
The prizes are subject to any taxes that may be required