demo code release

Author: akanazawa

.gitignore

@@ -1 +1,6 @@
-.DS_Store
+models/
+logs*/
+*.json
+*.DS_Store
+*.pyc
+forrelease.sh

README.md

@@ -1,8 +1,72 @@
 # End-to-end Recovery of Human Shape and Pose
 
 Angjoo Kanazawa, Michael J. Black, David W. Jacobs, Jitendra Malik
+CVPR 2018
 
 [Project Page](https://akanazawa.github.io/hmr/)
 ![Teaser Image](https://akanazawa.github.io/hmr/resources/images/teaser.png)
 
+### Requirements
+- Python 2.7
+- [TensorFlow](https://www.tensorflow.org/) tested on version 1.3
 
+### Installation
+
+#### Setup virtualenv
+```
+virtualenv venv_hmr
+source venv_hmr/bin/activate
+pip install -U pip
+deactivate
+source venv_hmr/bin/activate
+pip install -r requirements.txt
+```
+#### Install TensorFlow
+With GPU:
+```
+pip install tensorflow-gpu==1.3.0
+```
+Without GPU:
+```
+pip install tensorflow==1.3.0
+```
+
+### Demo
+
+1. Download the pre-trained models
+```
+wget https://people.eecs.berkeley.edu/~kanazawa/cachedir/hmr/models.tar.gz && tar -xf models.tar.gz
+```
+
+2. Run the demo
+```
+python -m demo --img_path data/coco1.png
+python -m demo --img_path data/im1954.jpg
+```
+
+On images that are not tightly cropped, you can run
+[openpose](https://github.com/CMU-Perceptual-Computing-Lab/openpose) and supply
+its output json (run it with `--write_json` option).
+When json_path is specified, the demo will compute the right scale and bbox center to run HMR:
+```
+python -m demo --img_path data/random.jpg --json_path data/random_keypoints.json
+```
+(The demo only runs on the most confident bounding box, see `src/util/openpose.py:get_bbox`)
+
+### Training code/data
+
+Coming soon.
+
+### Citation
+If you use this code for your research, please consider citing:
+```
+@inProceedings{kanazawaHMR18,
+  title={End-to-end Recovery of Human Shape and Pose},
+  author = {Angjoo Kanazawa
+  and Michael J. Black
+  and David W. Jacobs
+  and Jitendra Malik},
+  booktitle={Computer Vision and Pattern Regognition (CVPR)},
+  year={2018}
+}
+```

__init__.py

@@ -0,0 +1,135 @@
+"""
+Demo of HMR.
+
+Note that HMR requires the bounding box of the person in the image. The best performance is obtained when max length of the person in the image is roughly 150px. 
+
+When only the image path is supplied, it assumes that the image is centered on a person whose length is roughly 150px.
+Alternatively, you can supply output of the openpose to figure out the bbox and the right scale factor.
+
+Sample usage:
+
+# On images on a tightly cropped image around the person
+python -m demo --img_path data/im1963.jpg
+python -m demo --img_path data/coco1.png
+
+# On images, with openpose output
+python -m demo --img_path data/random.jpg --json_path data/random_keypoints.json
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import sys
+from absl import flags
+import numpy as np
+
+import skimage.io as io
+import tensorflow as tf
+
+from src.util import renderer as vis_util
+from src.util import image as img_util
+from src.util import openpose as op_util
+import src.config
+from src.RunModel import RunModel
+
+flags.DEFINE_string('img_path', 'data/im1963.jpg', 'Image to run')
+flags.DEFINE_string(
+    'json_path', None,
+    'If specified, uses the openpose output to crop the image.')
+
+
+def visualize(img, proc_param, joints, verts, cam):
+    """
+    Renders the result in original image coordinate frame.
+    """
+    cam_for_render, vert_shifted, joints_orig = vis_util.get_original(
+        proc_param, verts, cam, joints, img_size=img.shape[:2])
+
+    # Render results
+    skel_img = vis_util.draw_skeleton(img, joints_orig)
+    rend_img_overlay = renderer(
+        vert_shifted, cam=cam_for_render, img=img, do_alpha=True)
+    rend_img = renderer(
+        vert_shifted, cam=cam_for_render, img_size=img.shape[:2])
+    rend_img_vp1 = renderer.rotated(
+        vert_shifted, 60, cam=cam_for_render, img_size=img.shape[:2])
+    rend_img_vp2 = renderer.rotated(
+        vert_shifted, -60, cam=cam_for_render, img_size=img.shape[:2])
+
+    import matplotlib.pyplot as plt
+    # plt.ion()
+    plt.figure(1)
+    plt.clf()
+    plt.subplot(231)
+    plt.imshow(img)
+    plt.title('input')
+    plt.axis('off')
+    plt.subplot(232)
+    plt.imshow(skel_img)
+    plt.title('joint projection')
+    plt.axis('off')
+    plt.subplot(233)
+    plt.imshow(rend_img_overlay)
+    plt.title('3D Mesh overlay')
+    plt.axis('off')
+    plt.subplot(234)
+    plt.imshow(rend_img)
+    plt.title('3D mesh')
+    plt.axis('off')
+    plt.subplot(235)
+    plt.imshow(rend_img_vp1)
+    plt.title('diff vp')
+    plt.axis('off')
+    plt.subplot(236)
+    plt.imshow(rend_img_vp2)
+    plt.title('diff vp')
+    plt.axis('off')
+    plt.draw()
+    plt.show()
+    # import ipdb
+    # ipdb.set_trace()
+
+
+def preprocess_image(img_path, json_path=None):
+    img = io.imread(img_path)
+
+    if json_path is None:
+        scale = 1.
+        center = np.round(np.array(img.shape[:2]) / 2).astype(int)
+        # image center in (x,y)
+        center = center[::-1]
+    else:
+        scale, center = op_util.get_bbox(json_path)
+
+    crop, proc_param = img_util.scale_and_crop(img, scale, center,
+                                               config.img_size)
+
+    # Normalize image to [-1, 1]
+    crop = 2 * ((crop / 255.) - 0.5)
+
+    return crop, proc_param, img
+
+
+def main(img_path, json_path=None):
+    sess = tf.Session()
+    model = RunModel(config, sess=sess)
+
+    input_img, proc_param, img = preprocess_image(img_path, json_path)
+    # Add batch dimension: 1 x D x D x 3
+    input_img = np.expand_dims(input_img, 0)
+
+    joints, verts, cams, joints3d, theta = model.predict(
+        input_img, get_theta=True)
+
+    visualize(img, proc_param, joints[0], verts[0], cams[0])
+
+
+if __name__ == '__main__':
+    config = flags.FLAGS
+    config(sys.argv)
+
+    config.batch_size = 1
+
+    renderer = vis_util.SMPLRenderer(face_path=config.smpl_face_path)
+
+    main(config.img_path, config.json_path)

data/coco1.png

@@ -0,0 +1,5 @@
+models/
+logs*/
+*.json
+*.DS_Store
+*.pyc

data/coco2.png

@@ -0,0 +1,72 @@
+# End-to-end Recovery of Human Shape and Pose
+
+Angjoo Kanazawa, Michael J. Black, David W. Jacobs, Jitendra Malik
+CVPR 2018
+
+[Project Page](https://akanazawa.github.io/hmr/)
+![Teaser Image](https://akanazawa.github.io/hmr/resources/images/teaser.png)
+
+### Requirements
+- Python 2.7
+- [TensorFlow](https://www.tensorflow.org/) tested on version 1.3
+
+### Installation
+
+#### Setup virtualenv
+```
+virtualenv venv_hmr
+source venv_hmr/bin/activate
+pip install -U pip
+deactivate
+source venv_hmr/bin/activate
+pip install -r requirements.txt
+```
+#### Install TensorFlow
+With GPU:
+```
+pip install tensorflow-gpu==1.3.0
+```
+Without GPU:
+```
+pip install tensorflow==1.3.0
+```
+
+### Demo
+
+1. Download the pre-trained models
+```
+wget https://people.eecs.berkeley.edu/~kanazawa/cachedir/hmr/models.tar.gz && tar -xf models.tar.gz
+```
+
+2. Run the demo
+```
+python -m demo --img_path data/coco1.png
+python -m demo --img_path data/im1954.jpg
+```
+
+On images that are not tightly cropped, you can run
+[openpose](https://github.com/CMU-Perceptual-Computing-Lab/openpose) and supply
+its output json (run it with `--write_json` option).
+When json_path is specified, the demo will compute the right scale and bbox center to run HMR:
+```
+python -m demo --img_path data/random.jpg --json_path data/random_keypoints.json
+```
+(The demo only runs on the most confident bounding box, see `src/util/openpose.py:get_bbox`)
+
+### Training code/data
+
+Coming soon.
+
+### Citation
+If you use this code for your research, please consider citing:
+```
+@inProceedings{kanazawaHMR18,
+  title={End-to-end Recovery of Human Shape and Pose},
+  author = {Angjoo Kanazawa
+  and Michael J. Black
+  and David W. Jacobs
+  and Jitendra Malik},
+  booktitle={Computer Vision and Pattern Regognition (CVPR)},
+  year={2018}
+}
+```