Bachelor Thesis written by Kamaladdin Ahmadzada
CIMSI project combines two different deep neural network architecture itself. First one is Pix2Pix which belongs to Conditional Generative Adversarial Network family.Gets sketch file as a input and this network returns realistic-looking criminal photo. *This should be noted here that all these sketch files must be as much similar as to criminal.Otherwise confident rate drops.In second part of the network given input as a GAN result passes to Siamese net which is widely used for high accuracy one-shot learning technique.Siamese net is a good result of transfer learning in this project ResNet50 architecture used for classification but all this network's training performed by Oxford University. Pre-trained model can be found here .In vgg-face2 whole dataset can be found but it is quite large ~40GB that is why pre-trained model preferred .
It took ~4 hours to get reasonable results like images below.Hardware Specs:
- 1 Tesla T4, GPU
- Architecture: x86_64
- CPU op-mode(s): 32-bit, 64-bit
- Byte Order: Little Endian
- CPU(s): 2
- On-line CPU(s) list: 0,1
- Thread(s) per core: 2
- Core(s) per socket: 1
- Socket(s): 1
- NUMA node(s): 1
- Vendor ID: GenuineIntel
- CPU family: 6
- Model: 63
- Model name: Intel(R) Xeon(R) CPU @ 2.30GHz
- Stepping: 0
- CPU MHz: 2300.000
- BogoMIPS: 4600.00
- Hypervisor vendor: KVM
- Virtualization type: full
- L1d cache: 32K
- L1i cache: 32K
- L2 cache: 256K
- L3 cache: 46080K
- NUMA node0 CPU(s): 0,1
Here are some GAN result 1 :
