FASCODE

---

The FASCODE (FAShion COordination DatasEt / Fashion CODE) provided for Sub-Task 1 and 2 consists of 14,000 fashion image data. It includes bounding box coordinates (xmax, ymax, xmin, ymin) representing the position information of each garment and labeled information for three categories of emotional features (everydayness, sexiness, decoration) assigned to each item.
(Detailed information can be found in the dataset and codebook).

Please refer to the "Announcements and FAQs" board for frequently asked questions!

The "Announcements and FAQs" board is a participant support page created with Notion and can be accessed using Chrome, Edge, Mobile, Safari, etc. Please note that it is not compatible with Internet Explorer.

Evaluation Metrics: Averge Class-Specific Accuracy (ACSA)

---

ACSA represents the average accuracy of individual class classifications. It is calculated by obtaining the test results in the form of a Confusion Matrix and substituting them into the following formula:

Dataset

---

Develop an evaluation system/artificial intelligence during the competition period to classify imbalanced color categories within the fashion data using the FASCODE training and development sets. After constructing the model, submit the results to the leaderboard

- Training set: FASCODE dataset.
* Refer to the specific Sub-Task limitations (refer to the Restrictions above)
- Development set (Evaluation set) released (simultaneously with the training set)
- Provided data input/output and evaluation codes
- FASCODE encoding: EUC-KR
* Submission: 3 times a day, interval of 1hour between submissions


※ By submitting the model, it is assumed that you agree to calculate the test scores and share them on the leaderboard.

Download the dataset (link)
Download the baseline (link)
Download Validation Set (link)
Download Main Package List
Download Full Package List