Update README.md
Browse files
README.md
CHANGED
|
@@ -6,16 +6,11 @@ Official Pytorch code of [Instance-Conditioned GAN](https://arxiv.org/abs/2109.0
|
|
| 6 |
We provide a [Google Colab notebook](https://colab.research.google.com/github/facebookresearch/ic_gan/blob/main/inference/icgan_colab.ipynb) to generate images with IC-GAN and its class-conditional counter part. We also invite users to check out the [demo on Replicate](https://replicate.ai/arantxacasanova/ic_gan), courtesy of [Replicate](https://replicate.ai/home).
|
| 7 |
|
| 8 |
The figure below depicts two instances, unseen during training and downloaded from [Creative Commons search](https://search.creativecommons.org), and the generated images with IC-GAN and class-conditional IC-GAN when conditioning on the class "castle":
|
| 9 |
-
|
| 10 |
-
<img src="./figures/icgan_transfer_all_github.png?raw=true">
|
| 11 |
-
</p>
|
| 12 |
|
| 13 |
Additionally, and inspired by [this Colab](https://colab.research.google.com/github/eyaler/clip_biggan/blob/main/ClipBigGAN.ipynb), we provide the funcionality in the same Colab notebook to guide generations with text captions, using the [CLIP model](https://github.com/openai/CLIP).
|
| 14 |
As an example, the following Figure shows three instance conditionings and a text caption (top), followed by the resulting generated images with IC-GAN (bottom), when optimizing the noise vector following CLIP's gradient for 100 iterations.
|
| 15 |
-
|
| 16 |
-
<img src="./figures/icgan_clip.png?raw=true">
|
| 17 |
-
</p>
|
| 18 |
-
|
| 19 |
|
| 20 |
*Credit for the three instance conditionings, from left to right, that were modified with a resize and central crop:* [1: "Landscape in Bavaria" by shining.darkness, licensed under CC BY 2.0](https://search.creativecommons.org/photos/92ef279c-4469-49a5-aa4b-48ad746f2dc4), [2: "Fantasy Landscape - slolsss" by Douglas Tofoli is marked with CC PDM 1.0](https://search.creativecommons.org/photos/13646adc-f1df-437a-a0dd-8223452ee46c), [3: "How to Draw Landscapes Simply" by Kuwagata Keisai is marked with CC0 1.0](https://search.creativecommons.org/photos/2ab9c3b7-de99-4536-81ed-604ee988bd5f)
|
| 21 |
|
|
|
|
| 6 |
We provide a [Google Colab notebook](https://colab.research.google.com/github/facebookresearch/ic_gan/blob/main/inference/icgan_colab.ipynb) to generate images with IC-GAN and its class-conditional counter part. We also invite users to check out the [demo on Replicate](https://replicate.ai/arantxacasanova/ic_gan), courtesy of [Replicate](https://replicate.ai/home).
|
| 7 |
|
| 8 |
The figure below depicts two instances, unseen during training and downloaded from [Creative Commons search](https://search.creativecommons.org), and the generated images with IC-GAN and class-conditional IC-GAN when conditioning on the class "castle":
|
| 9 |
+
|
|
|
|
|
|
|
| 10 |
|
| 11 |
Additionally, and inspired by [this Colab](https://colab.research.google.com/github/eyaler/clip_biggan/blob/main/ClipBigGAN.ipynb), we provide the funcionality in the same Colab notebook to guide generations with text captions, using the [CLIP model](https://github.com/openai/CLIP).
|
| 12 |
As an example, the following Figure shows three instance conditionings and a text caption (top), followed by the resulting generated images with IC-GAN (bottom), when optimizing the noise vector following CLIP's gradient for 100 iterations.
|
| 13 |
+
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
*Credit for the three instance conditionings, from left to right, that were modified with a resize and central crop:* [1: "Landscape in Bavaria" by shining.darkness, licensed under CC BY 2.0](https://search.creativecommons.org/photos/92ef279c-4469-49a5-aa4b-48ad746f2dc4), [2: "Fantasy Landscape - slolsss" by Douglas Tofoli is marked with CC PDM 1.0](https://search.creativecommons.org/photos/13646adc-f1df-437a-a0dd-8223452ee46c), [3: "How to Draw Landscapes Simply" by Kuwagata Keisai is marked with CC0 1.0](https://search.creativecommons.org/photos/2ab9c3b7-de99-4536-81ed-604ee988bd5f)
|
| 16 |
|