Deepfake Detection - Transfer Learning Imagenet models to detect deepfaked images
Created on May 10|Last edited on June 9
Comment
Classifier Runs
meta
["-f","/root/.local/share/jupyter/runtime/kernel-d6b9db77-cd59-4763-a61b-2e2643fb537c.json"]
["-f","/root/.local/share/jupyter/runtime/kernel-d6b9db77-cd59-4763-a61b-2e2643fb537c.json"]
["-f","/root/.local/share/jupyter/runtime/kernel-d6b9db77-cd59-4763-a61b-2e2643fb537c.json"]
["-f","/root/.local/share/jupyter/runtime/kernel-d6b9db77-cd59-4763-a61b-2e2643fb537c.json"]
["-f","/root/.local/share/jupyter/runtime/kernel-d6b9db77-cd59-4763-a61b-2e2643fb537c.json"]
["-f","/home/ishaanmalhi/.local/share/jupyter/runtime/kernel-7291a5a1-24c5-40cd-9081-f20311166331.json"]
["-f","/home/ishaanmalhi/.local/share/jupyter/runtime/kernel-b3fc557a-7331-4840-a804-4ecca34c5e9c.json"]
["-f","/home/ishaanmalhi/.local/share/jupyter/runtime/kernel-b3fc557a-7331-4840-a804-4ecca34c5e9c.json"]
["-f","/home/ishaanmalhi/.local/share/jupyter/runtime/kernel-0acf5350-b605-4605-a749-afe209aed7bb.json"]
["-f","/home/ishaanmalhi/.local/share/jupyter/runtime/kernel-37112f9c-2f74-4665-9f8a-87448bf54c1b.json"]
8
8
8
8
8
12
6
6
6
6
Quadro P5000
Quadro P5000
Quadro P5000
Quadro P5000
Quadro P5000
Tesla K80
Tesla K80
Tesla K80
Tesla K80
Tesla K80
1
1
1
1
1
2
1
1
1
1
Final 8 epoch run
Finetune and train} entire network after transfer learning on deepfake images.
Samples all face crops and train on full dataset
Trained last 5 layers of efficientnet b7
Final resnext101 run
Final model for widerresnext
Linux-4.15.0-72-generic-x86_64-with-Ubuntu-18.04-bionic
Linux-4.15.0-72-generic-x86_64-with-Ubuntu-18.04-bionic
Linux-4.15.0-72-generic-x86_64-with-Ubuntu-18.04-bionic
Linux-4.15.0-72-generic-x86_64-with-Ubuntu-18.04-bionic
Linux-4.15.0-72-generic-x86_64-with-Ubuntu-18.04-bionic
Linux-5.3.0-1020-azure-x86_64-with-debian-buster-sid
Linux-5.3.0-1020-azure-x86_64-with-debian-buster-sid
Linux-5.3.0-1020-azure-x86_64-with-debian-buster-sid
Linux-5.3.0-1020-azure-x86_64-with-debian-buster-sid
Linux-5.3.0-1020-azure-x86_64-with-debian-buster-sid
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
<python with no main file>
3.6.9
3.6.9
3.6.9
3.6.9
3.6.9
3.7.6
3.7.6
3.7.6
3.7.6
3.7.6
33m 42s
41m 20s
10m 16s
14m 36s
3s
4h 31m 33s
1h 52m 50s
29m 7s
3h 19m 6s
2h 25m 38s
config
2
2
2
2
-
-
-
-
-
-
0.001
0.0001
0.01
0.001
-
-
-
-
-
-
-
-
efficientnet-affine-transforms-v2
-
-
-
-
-
-
-
Compose(
ColorJitter(brightness=[0.5, 1.5], contrast=[0.5, 1.5], saturation=[0.5, 1.5], hue=None)
RandomHorizontalFlip(p=0.5)
RandomAffine(degrees=(-2, 2), translate=(0.3, 0.3))
CenterCrop(size=(149, 149))
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
-
-
-
-
-
-
-
-
-
summary(8 collapsed)
Run set
22
Add a comment