From 7d32e838e9d16f8f2328276ba8b5a2dad379a4f1 Mon Sep 17 00:00:00 2001
From: Bharat-mtr <bharatagmtr.999@gmail.com>
Date: Sat, 15 Oct 2022 20:56:54 +0530
Subject: [PATCH 1/2] Added readme

---
 .../README.md                                 | 81 +++++++++++++++++++
 1 file changed, 81 insertions(+)
 create mode 100644 Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md

diff --git a/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md b/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md
new file mode 100644
index 0000000..131efe3
--- /dev/null
+++ b/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md	
@@ -0,0 +1,81 @@
+# Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars
+
+A project on the speed up of one-shot adversarially trained human pose to image translation models for mobile devices.
+
+<img src="https://saic-violet.github.io/bilayer-model/assets/teaser.png"/>
+
+## Installation
+
+* Python 3.7
+* Pytorch 1.3 or higher
+* Apex (is required only for training, needs to be built from https://github.com/NVIDIA/apex)
+* Face-alignment (https://github.com/1adrianb/face-alignment)
+* Other packages are in requirements.txt
+* Download pretrained_weights and runs from https://dagshub.com/Bharat-mtr/bilayer-model/src/master/model 
+
+## Inference API usage
+
+```python
+import argparse
+from infer import InferenceWrapper
+
+args_dict = {
+    'project_dir': '.',
+    'init_experiment_dir': './runs/vc2-hq_adrianb_paper_main',
+    'init_networks': 'identity_embedder, texture_generator, keypoints_embedder, inference_generator',
+    'init_which_epoch': '2225',
+    'num_gpus': 1,
+    'experiment_name': 'vc2-hq_adrianb_paper_enhancer',
+    'which_epoch': '1225',
+    'spn_networks': 'identity_embedder, texture_generator, keypoints_embedder, inference_generator, texture_enhancer',
+    'enh_apply_masks': False,
+    'inf_apply_masks': False}
+
+# Initialization
+module = InferenceWrapper(args_dict)
+
+# Input data for intiialization and inference
+data_dict = {
+    'source_imgs': ..., # Size: H x W x 3, type: NumPy RGB uint8 image
+    'target_imgs': ..., # Size: NUM_FRAMES x H x W x 3, type: NumPy RGB uint8 images
+}
+
+# Inference
+data_dict = module(data_dict)
+
+# Outputs (images are in [-1, 1] range, segmentation masks -- in [0, 1])
+imgs = data_dict['pred_enh_target_imgs']
+segs = data_dict['pred_target_segs']
+```
+
+For a concrete inference example, please refer to examples/inference.ipynb.
+
+## Training
+
+The example training scripts are in the scripts folder. The base model is trained first, the texture enhancer is trained afterwards. In order to reproduce the results from the paper, 8 GPUs with at least 24 GBs of memory are required, since batch normalization layers may be sensitive to the batch size.
+
+## Datasets
+
+Supported datasets should have the same structure as VoxCeleb2 (http://www.robots.ox.ac.uk/~vgg/data/voxceleb/vox2.html) dataset:
+
+```DATA_ROOT/[imgs, keypoints, segs]/[train, test]/PERSON_ID/VIDEO_ID/SEQUENCE_ID/FRAME_NUM[.jpg, .npy, .png]```
+
+Please refer to the link above for more details.
+
+Additionally, all training data must be annotated with keypoints obtained using face-alignment (or any other keypoints detection) library before training. Annotation with segmentation masks is optional, yet it significantly improves the performance of the method.
+
+## Links
+
+- Project page: https://saic-violet.github.io/bilayer-model
+- ArXiv: https://arxiv.org/abs/2008.10174
+- YouTube: https://youtu.be/54tji11VhOI
+
+## Citation
+```
+@InProceedings{Zakharov20,
+  author={Zakharov, Egor and Ivakhnenko, Aleksei and Shysheya, Aliaksandra and Lempitsky, Victor},
+  title={Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars},
+  booktitle = {European Conference of Computer vision (ECCV)},
+  month = {August},
+  year = {2020}}
+```

From fc71ec98bcbf66f526b7d366bc6c40fc3ba9c95e Mon Sep 17 00:00:00 2001
From: Bharat-mtr <62692986+Bharat-mtr@users.noreply.github.com>
Date: Sat, 15 Oct 2022 21:01:22 +0530
Subject: [PATCH 2/2] Update README.md

---
 .../README.md                                                  | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md b/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md
index 131efe3..14aee59 100644
--- a/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md	
+++ b/Fast Bi-layer Neural Synthesis of One-Shot Realistic Head Avatars/README.md	
@@ -4,6 +4,9 @@ A project on the speed up of one-shot adversarially trained human pose to image
 
 <img src="https://saic-violet.github.io/bilayer-model/assets/teaser.png"/>
 
+
+[DagsHub Repository](https://dagshub.com/Bharat-mtr/bilayer-model)
+
 ## Installation
 
 * Python 3.7