test_single_triangle_camera_fisheye.py

# Tensorflow by default allocates all GPU memory, leaving very little for rendering.
# We set the environment variable TF_FORCE_GPU_ALLOW_GROWTH to true to enforce on demand
# memory allocation to reduce page faults.
import os
os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'
import tensorflow as tf
tf.compat.v1.enable_eager_execution()
import pyredner_tensorflow as pyredner

# Optimize fisheye camera parameters of a single triangle rendering

# Use GPU if available
pyredner.set_use_gpu(tf.test.is_gpu_available(cuda_only=True, min_cuda_compute_capability=None))

# Set up the scene
with tf.device('/device:cpu:' + str(pyredner.get_cpu_device_id())):
    position = tf.Variable([0.0, 0.0, -1.0], dtype=tf.float32)
    look_at = tf.Variable([0.0, 0.0, 0.0], dtype=tf.float32)
    up = tf.Variable([0.0, 1.0, 0.0], dtype=tf.float32)
    fov = tf.Variable([45.0], dtype=tf.float32)
    clip_near = 1e-2
    resolution = (256, 256)
    cam = pyredner.Camera(position = position,
                          look_at = look_at,
                          up = up,
                          fov = fov,
                          clip_near = clip_near,
                          resolution = resolution,
                          fisheye = True)

with tf.device(pyredner.get_device_name()):
    mat_grey = pyredner.Material(diffuse_reflectance = \
        tf.Variable([0.5, 0.5, 0.5], dtype=tf.float32))
    materials = [mat_grey]
    vertices = tf.Variable([[-1.7,1.0,0.0], [1.0,1.0,0.0], [-0.5,-1.0,0.0]],
        dtype=tf.float32)
    indices = tf.constant([[0, 1, 2]], dtype=tf.int32)
    shape_triangle = pyredner.Shape(vertices, indices, 0)
    light_vertices = tf.Variable([[-1.0,-1.0,-9.0],[1.0,-1.0,-9.0],[-1.0,1.0,-9.0],[1.0,1.0,-9.0]],
        dtype=tf.float32)
    light_indices = tf.constant([[0,1,2],[1,3,2]], dtype=tf.int32)
    shape_light = pyredner.Shape(light_vertices, light_indices, 0)
shapes = [shape_triangle, shape_light]

with tf.device('/device:cpu:' + str(pyredner.get_cpu_device_id())):
    light_intensity = tf.Variable([30.0,30.0,30.0], dtype=tf.float32)
light = pyredner.AreaLight(1, light_intensity)
area_lights = [light]

scene = pyredner.Scene(cam, shapes, materials, area_lights)
scene_args = pyredner.serialize_scene(
    scene = scene,
    num_samples = 16,
    max_bounces = 1)

# Render our target
img = pyredner.render(0, *scene_args)
pyredner.imwrite(img, 'results/test_single_triangle_camera_fisheye/target.exr')
pyredner.imwrite(img, 'results/test_single_triangle_camera_fisheye/target.png')
target = pyredner.imread('results/test_single_triangle_camera_fisheye/target.exr')

# Perturb the scene, this is our initial guess
with tf.device('/device:cpu:' + str(pyredner.get_cpu_device_id())):
    position = tf.Variable([0.5, -0.5, -3.0], trainable=True)
scene.camera = pyredner.Camera(position = position,
                               look_at = look_at,
                               up = up,
                               fov = fov,
                               clip_near = clip_near,
                               resolution = resolution,
                               fisheye = True)
scene_args = pyredner.serialize_scene(
    scene = scene,
    num_samples = 16,
    max_bounces = 1)
# Render the initial guess
img = pyredner.render(1, *scene_args)
pyredner.imwrite(img, 'results/test_single_triangle_camera_fisheye/init.png')
diff = tf.abs(target - img)
pyredner.imwrite(diff, 'results/test_single_triangle_camera_fisheye/init_diff.png')

# Optimize for camera pose
optimizer = tf.compat.v1.train.AdamOptimizer(2e-2)
for t in range(200):
    print('iteration:', t)
    
    with tf.GradientTape() as tape:
        # Need to rerun the Camera constructor for PyTorch autodiff to compute the derivatives
        scene.camera = pyredner.Camera(position   = position,
                                       look_at    = look_at,
                                       up         = up,
                                       fov        = fov,
                                       clip_near  = clip_near,
                                       resolution = resolution,
                                       fisheye    = True)
        scene_args = pyredner.serialize_scene(
            scene = scene,
            num_samples = 4,
            max_bounces = 1)
        img = pyredner.render(t+1, *scene_args)
        pyredner.imwrite(img, 'results/test_single_triangle_camera_fisheye/iter_{}.png'.format(t))
        loss = tf.reduce_sum(tf.square(img - target))
    
    print('loss:', loss)

    grads = tape.gradient(loss, [position])

    optimizer.apply_gradients(zip(grads, [position]))

    print('d_position:', grads[0])
    print('position:', position)

scene_args = pyredner.serialize_scene(
    scene = scene,
    num_samples = 16,
    max_bounces = 1)
img = pyredner.render(202, *scene_args)
pyredner.imwrite(img, 'results/test_single_triangle_camera_fisheye/final.exr')
pyredner.imwrite(img, 'results/test_single_triangle_camera_fisheye/final.png')
pyredner.imwrite(tf.abs(target - img), 'results/test_single_triangle_camera_fisheye/final_diff.png')

from subprocess import call
call(["ffmpeg", "-framerate", "24", "-i",
    "results/test_single_triangle_camera_fisheye/iter_%d.png", "-vb", "20M",
    "results/test_single_triangle_camera_fisheye/out.mp4"])