| 1 | 0.1 | 0 | 1000 | 100
| --- | --- | --- | --- | ---
|
|
|
|
|
|
From above L1 weight as 0.01 gives the best result.
| 9 | 1 | 0.1 | 0.01 | 0.001
| --- | --- | --- | --- | ---
|
|
|
|
|
|
From above Perceptual loss weight as 10 gives the best result.
| VanillaGAN | StyleGAN
| --- | --- |
|
| 
StyleGAN gives better results.
| z | w | w+
| --- | --- | --- |
|
|
| 
w+ captures better color details.
Inference Time: All models take approximately 18-20 seconds to run.
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
|
| --- | --- | --- |
|
|
| 
Findings from above:
Results:
| 1 | 2
| --- | --- |
|
| 
| Sketch | 100 | 250 | 500
| --- | --- | --- | ---
|
|
|
|
|
More noise helps make the generated image farther away from the sketch.
| Sketch | 10 | 15 | 30 | 60
| --- | --- | --- | --- | ---
|
|
|
|
|
|
Increasing the strength of classifier-guidance makes the cat more grumpy i.e. more aligned to the prompt.
| BW1 | BW2
| --- | ---
|
|
|