The 1st FR FRONTIER: Color detection in fashion images
- 2017/07/14 (Closed)
- Number of Participants
1st place 1,000,000 JPY
2nd place 300,000 JPY
3rd place 100,000 JPY
Special Prize 100,000 JPY
- Host / Sponsor
This competition is intended to discover the brilliant talents.
You must accept the agreement to enter this competition.
Changing clothes. Changing conventional wisdom. Change the world.
The Fast Retailing Group Mission is to create truly great clothing with new and unique value, and to enable people all over the world to experience the joy, happiness and satisfaction of wearing such great clothes. We work to delight customers in innovative ways, working with talented people and leveraging great ideas from all around the world.
FR FRONTIER is a new international competition sponsored by Fast Retailing. We are looking for people who are interested to challenge the future and reach a new frontier. We invite you to submit creative solutions to help solve today's business challenges, using your business planning skills, technical strengths and creativity.
Please also see here for the details of this competition.
ChallengeColor detection in fashion images
When developing new products, we analyze mass consumer trends after classifying the color of apparel from various fashion photographs. We are employing new machine learning algorithms to achieve this classification faster and more accurate.
Conditions for participation
- Undergraduate or graduate students or working professionals
- Individuals and teams of no more than 5 members
* Please use the team leader's account to submit predictions.
- Ability to submit materials, presentations, and handle Q&A in Japanese or English
- Original proposal, not used before in other competitions
- Ability to attend the Awards Ceremony in Tokyo (only for winning candidates)
* Fees for attending the ceremony will be paid(Transportation and hotel expenses). Visa and travel must be arranged by participants, and the venue will be informed separately to the Awards Ceremony participants.
Sign up for DeepAnalytics to enter this competition.
Friday, April 28, 2017 - Competition Launch Date
Friday, July 14, 2017 - Prediction Submission Deadline
Monday, July 17, 2017 - Source Code Submission Deadline (only for the winning cadidates)
Friday, July 28, 2017 - Announcement of Initial Results
Monday, August 28, 2017 - Final Presentation and Awards Ceremony
NOTE: All deadlines are at 11:59 PM JST on the corresponding day unless otherwise noted.
Creating a team
Please follow these procedures to create a team.
1. All of the team members must sign up for DeepAnalytics.
2. All of the team members must agree to the Opt-in of this competition.
3. The team leader must send us the list of member's email addresses via Contact Us form.
One account per individual
You may have one account per perticipant.
If you enter as a team, only team leader's account is allowed to submit predictions.
No private sharing outside teams
Privately sharing code or data outside of teams is not permitted.
No manual labeling
You must not submit files with predictions labeled manually.
No data fabrication / falsification
You must not modify labels in the training set to train your model.
Do not edit image files
You must not edit image files with using softwares (i.e. removing background from image).
No use of external data
You must not train your model with using either external data or test data set.
Using external trained models and third party libraries
You may use the external trained models only if they are open source models (source code is provided, anyone can access, and license free).
In using external trained models, please clarify their referring source (i.e. paper link).
You may use the libraries only if they are free to use (i.e. tensorflow, chainer, and so on).
Your challenge in this competition is to detect the color on clothing featured in fashion images and classify it into 24 categories.
Please see the Download Data page for the details of submission format.
We use the metric "Balanced Accuracy" to score the submissions.
The higher submission score you get, the higher rank you will be placed in.
1. If your submission has the same score with the other participant's submission, the rank will be placed on a first-submit.
2. After the competition finished, we determine the final rankings by evaluating your submissions with different data with the one during competition.
3. We will ask you to submit the following items when determining winners.
a. The code and document of your model to reproduce the result.
b. Contribution of each feature in your classification model.
c. Execution environment (OS version, the software and modeling method that you used).
d. In your modeling with using random numbers (i.e. Random forest), please fix the seed value so that the model can be reproduced by third parties.
4. When the person or its submitted model falls under any of the following, he will lose his eligibility to receive the prize.
a. The person does not respond to the contact from us within the prescribed period.
b. Unable to reproduce the submission predictions during the competition.
c. The model lacks robustness (i.e. There are relatively more white t-shirts than yellow ones).
Guideline for modeling
Please follow the guideline below for the purpose of smooth communication upon the announcement of winning candidates.
1. Please separate your source code into 3 parts: preprocessing, training, and predicting.
a. Let's say you are coding in Python, you shall respectively code:
- preprocess.py for preprocessing
- train.py for training
- predict.py for making predictions
b. Use any languages to code your algorithms (i.e. C, Java, Lua, R, as such)
c. Sample implementation code is available. Please refer to sample_code.zip located see this page
2. Please submit your trained model as well, not only its source code.
a. We confirm if the trained model can reproduce your score on the leaderboard with using test set
3. Before submitting your code, make sure your latest submitted predictions and your local predictions match.
a. Please check the path in the test data or trained model
b. Confirm the command to execute the codes/scripts
c. Confirm if there are errors within the script (i.e. predict.py)
d. We are much appreciated if you provide your code's execution time with us
Final presentation venue
Date: Monday, August 28, 2017
Location: Fast Retailing co., ltd. Ariake Office
Address: 6F UNIQLO CITY TOKYO, 1-6-7 Ariake, Koto-ku, Tokyo 135-0063, Japan
Procedure after Awards Ceremony
Please proceed with the neccesary procedures after Awards Ceremony (i.e. recieving prize money) with FAST RETAILING CO., LTD..
|23||Chen Zitian||0.67897||19||2017/07/14 22:35|
|24||ghalib ahmed||0.67779||21||2017/07/14 22:37|
|26||悲哀なる刃 †籠の中に囚われし加藤†||0.67671||19||2017/07/14 10:01|
|39||Ghalib Ahmed Tahir||0.66526||4||2017/07/14 23:53|
|42||Mustafa Ihsan||0.66159||8||2017/07/14 21:45|
|44||8 Café||0.65853||20||2017/07/08 22:46|