Welcome to Curio Face Swap AI Challenge!

NOTE: The dataset has been updated, please review again. The task is the same still.

All participants of the challenge are requested to submit their resumes, even if no submission done yet to: [email protected]

We have a developed a computer vision based 'face swap' app. We are looking for Computer Vision/Deep Learning enthusiasts to join a very passionate computer vision based product development team.

Specifically, we are looking for :

• A deep learning engineer who is familiar with GANs (eg. StyleGAN, conditional GANs, image2image translations, vid2vid), Autoregressive models, Flow models is a huge plus.
• A deep learning engineer who is familiar with 3D face Rendering, meshlabs, mesh topology, Blender tool, 3D Facial Landmark Extraction is a huge plus.
• Must have experience in developing GAN models such as deep fake, face-tempering, face-swapping
• Also, ideally looking for someone who is excellent at debugging deep learning models and is an expert at implementing models from scratch, using cloud infrastructure to train and deploy these models

The dataset consists of two directories:

1. images
2. videos

The aim of this challenge is to swap the faces provided in the images directory with the faces detected in videos present in the video directory. Each image has a name prefix of the video it has to be swapped in. The participants will need to submit the newly generated videos with the swapped faces as part of their submission.

NOTE: To know the exact submission file requirements and format, please check out "Submission Guidelines". Please read the guidelines carefully or your submission may be invalidated.

*************************************

Submission Evaluation: Manual

Submission Grade Time: 24 Hrs - 36 Hrs.

The results will be declared when ALL the submissions have been graded.

*************************************

NOTE: Please read the guidelines very carefully to understand the final structure of your submission file (.zip). Also, if you are using a Windows system, do not use WINRAR to zip the files, WinZip or UNIX based zip creators would be required.

The directory structure of the dataset is as follows:

1. images
2. videos

The images directory consists of 14 images.

The videos directory consists of 17 videos.

The aim is to swap the faces of the images provided in the image directory onto faces of videos provided in the videos directory.

1. 14 images and 17 videos can be processed into 51 new videos with each video processed with 3 images out of the 14 of your choice. Using different images for different videos is preferred.

The final submission (.zip) file should have the following files:

1. Processed videos (.mp4 format)
2. The Jupyter notebook/Colab Notebook used for training the model
3. Trained model and complete code directory.
4. Instructions to test the code, along with requirements (instructions.txt, requirements.txt)
5. A detailed report of your approach, what methods you tested. Also mention the methods which did not generate good results for you. (report.pdf)

If the zip submission format does not match, your solution WILL get invalidated.

How to make a submission?

1. Click on "My Submission"
2. On the next page, click on "+ New Submission"
3. Upload your CSV in the next page and click on "Submit for Review"

Please note:

1. You must submit your zip file by uploading the zip in the "My Submissions" section of this challenge.
2. If there is any error in the submission, your final score will be marked as 0 or invalid..

Judgement

1. Evaluation based on Video Results, Code Execution, Methodology given in the detailed report.
2. The solution will be graded out of 100.
3. If the scores are tied, the person reaching the score FIRST will get the better rank.

Rules

1. We are using automatic anti-cheat system. Our system flags submission which have low trust scores. Manually modifying your submission CSV (via comparison of any form etc.), using image comparison techniques (pixel matching, file size matching etc,), in any form, can lead to your disqualification without any notice.
2. The submission should be in a proper format as described by "Submission Guidelines".
3. Late submission will not be accepted beyond provided deadline (Indian Standard Time).
4. Confidentiality should be strictly maintained.

Rewards:

1. Participants will be invited for an interview based on score.
2. The participants with a successful submission will get a certificate of participation.
3. Top 3 participants will receive a permanent place in Dockship's "Hall of Fame".