The Effort Post
Welcome to another week of “becoming a nerd🤓” by doing not-so-nerdy activities. This past week I completely procrastinated (typical me) & didn’t finish what I am supposed to, which is giving the best narration of dissertation progress & my opinions about it.
As the week progressed, I started drafting my research questions on EfficientNets & the more I read, the more I got interested. Yeah, it doesn’t happen very often. 😉
The first question (🥁💡) I formed was just out of the basic curiosity of how well both the machine learning models will do when compared against each other as is. Right? Just the base level of difference & base level of performance ability on deepfake detection. My literature-ry version is something like this:
Is there a difference in the effectiveness of the state-of-the-art CNN deepfake detection techniques of EfficientNet-B4 and EfficientNet-B7 when untrained on any datasets and validated on the same benchmark dataset?
Do you have a KOOL Visual for this? OFC 😎
You must be wondering what the heck is a benchmark dataset. So it’s nothing but a combination of datasets that focuses on various types of image & video manipulation content.
I generated this benchmark dataset based on the technique of each dataset. It comprises 1,692 videos from Celeb-DF (Li et al., 2019), 30,000 videos from DFDC (Facebook, 2020), and 1,000 images from the FaceForensics++ benchmark dataset (Rössler et al., 2019) to equal 32,692 validation data points, which are not used as part of the training dataset for the EfficientNet-B4 and EfficientNet-B7 models.
FANCY + GENIUS + FREE DATA 🙌🏼
I have two fantastic things figured out, my research questions & my deepfake datasets. As we are comparing two machine learning models against fixed set data points, that means we have a quantitative research approach.
Believe me, for my selective extroverted personality, this is significantly better than a qualitative approach! I would rather question the science behind it than ask people to take my survey.
(Haha.. if you know you know 🤭😁)
The Journey doesn’t end here! But, it starts at Mega speed. So much to explore & so much to do (Oh hey!! ADHD😅 how are you doing?)
Yuhooo you made it to my (Jasmin Bharadiya) Journey so far!!
