search

NVlabs / Deep_Object_Pose

Deep Object Pose Estimation (DOPE) – ROS inference (CoRL 2018)
Other
1.03k stars 287 forks source link

Invalid Syntax error during synthetic dataset generation #290

Closed ArghyaChatterjee closed 1 year ago

ArghyaChatterjee commented 1 year ago

Hello, I am trying to generate synthetic dataset using nvisii. Here is how I am generating a single file:

arghya@arghya-Pulse-GL66-12UEK:~/dope_ws/src/dope/scripts/nvisii_data_gen$ python3 single_video_pybullet.py --nb_frames 1 --scale 0.01
pybullet build time: May 20 2022 19:44:17
folder output/dataset// exists
Warning, create_perspective_from_fov is deprecated and will be removed in a subsequent release. Please switch to create_from_fov.
00000/00001

But when I try the bulk datageneration pipeline, I get the following error:

arghya@arghya-Pulse-GL66-12UEK:~/dope_ws/src/dope/scripts/nvisii_data_gen$ python3 generate_dataset.py 
  File "single_video_pybullet.py", line 156
    if os.path.isdir(f'output/{opt.outf}'):
                                        ^
SyntaxError: invalid syntax
  File "single_video_pybullet.py", line 156
    if os.path.isdir(f'output/{opt.outf}'):
                                        ^
.....
.....
98 times ...
.....
SyntaxError: invalid syntax
  File "single_video_pybullet.py", line 156
    if os.path.isdir(f'output/{opt.outf}'):
                                        ^
SyntaxError: invalid syntax

Here is my generate dataset.py looks like:

#!/usr/bin/env python3

import os
import cv2
import glob
import math
import time
import nvisii as visii
import numpy as np
import pybullet as p
import random
import argparse
import colorsys
import subprocess
from math import acos
from math import sqrt
from math import pi

from utils import *

parser = argparse.ArgumentParser()

parser.add_argument(
    '--spp',
    default=800,
    type=int,
    help = "number of sample per pixel, higher the more costly"
)
parser.add_argument(
    '--width',
    default=500,
    type=int,
    help = 'image output width'
)
parser.add_argument(
    '--height',
    default=500,
    type=int,
    help = 'image output height'
)
# TODO: change for an array
parser.add_argument(
    '--objs_folder_distrators',
    default='google_scanned_models/',
    help = "object to load folder"
)
parser.add_argument(
    '--objs_folder',
    default='models/',
    help = "object to load folder"
)
parser.add_argument(
    '--path_single_obj',
    default=None,
    help='If you have a single obj file, path to the obj directly.'
)
parser.add_argument(
    '--scale',
    default=1,
    type=float,
    help='Specify the scale of the target object(s). If the obj mesh is in '
         'meters -> scale=1; if it is in cm -> scale=0.01.'
)

parser.add_argument(
    '--skyboxes_folder',
    default='dome_hdri_haven/',
    help = "dome light hdr"
)
parser.add_argument(
    '--nb_objects',
    default=1,
    type = int,
    help = "how many objects"
)
parser.add_argument(
    '--nb_distractors',
    default=7,
    help = "how many objects"
)
parser.add_argument(
    '--nb_frames',
    default=2000,
    help = "how many frames to save"
)
parser.add_argument(
    '--skip_frame',
    default=100,
    type=int,
    help = "how many frames to skip"
)
parser.add_argument(
    '--noise',
    action='store_true',
    default=False,
    help = "if added the output of the ray tracing is not sent to optix's denoiser"
)
parser.add_argument(
    '--outf',
    default='dataset/',
    help = "output filename inside output/"
)
parser.add_argument('--seed',
    default = None,
    help = 'seed for random selection'
)

parser.add_argument(
    '--interactive',
    action='store_true',
    default=False,
    help = "make the renderer in its window"
)

parser.add_argument(
    '--motionblur',
    action='store_true',
    default=False,
    help = "use motion blur to generate images"
)

parser.add_argument(
    '--box_size',
    default=0.5,
    type=float,
    help = "make the object movement easier"
)

parser.add_argument(
    '--focal-length',
    default=None,
    type=float,
    help = "focal length of the camera"
)

parser.add_argument(
    '--visibility-fraction',
    action='store_true',
    default=False,
    help = "Compute the fraction of visible pixels and store it in the "
           "`visibility` field of the json output. Without this argument, "
           "`visibility` is always set to 1. Slows down rendering by about "
           "50 %%, depending on the number of visible objects."
)

parser.add_argument(
    '--debug',
    action='store_true',
    default=False,
    help="Render the cuboid corners as small spheres. Only for debugging purposes, do not use for training!"
)

opt = parser.parse_args()

if os.path.isdir(f'output/{opt.outf}'):
    print(f'folder output/{opt.outf}/ exists')
else:
    os.makedirs(f'output/{opt.outf}')
    print(f'created folder output/{opt.outf}/')

opt.outf = f'output/{opt.outf}'

if not opt.seed is None:
    random.seed(int(opt.seed))

# # # # # # # # # # # # # # # # # # # # # # # # #

# # # # # # # # # # # # # # # # # # # # # # # # #

visii.initialize(headless = not opt.interactive)

if not opt.motionblur:
    visii.sample_time_interval((1,1))

visii.sample_pixel_area(
    x_sample_interval = (.5,.5),
    y_sample_interval = (.5, .5))

# visii.set_max_bounce_depth(1)

if not opt.noise:
    visii.enable_denoiser()

if opt.focal_length:
    camera = visii.entity.create(
        name = "camera",
        transform = visii.transform.create("camera"),
        camera = visii.camera.create_from_intrinsics(
            name = "camera",
            fx=opt.focal_length,
            fy=opt.focal_length,
            cx=(opt.width / 2),
            cy=(opt.height / 2),
            width=opt.width,
            height=opt.height
        )
    )
else:
    camera = visii.entity.create(
        name = "camera",
        transform = visii.transform.create("camera"),
        camera = visii.camera.create_perspective_from_fov(
            name = "camera",
            field_of_view = 0.785398,
            aspect = float(opt.width)/float(opt.height)
        )
    )

# data structure
random_camera_movement = {
    'at':visii.vec3(1,0,0),
    'up':visii.vec3(0,0,1),
    'eye':visii.vec3(0,0,0)
}

camera.get_transform().look_at(
    at = random_camera_movement['at'], # look at (world coordinate)
    up = random_camera_movement['up'], # up vector
    eye = random_camera_movement['eye'],
)

visii.set_camera_entity(camera)

# # # # # # # # # # # # # # # # # # # # # # # # #

# lets turn off the ambiant lights
# load a random skybox
skyboxes = glob.glob(f'{opt.skyboxes_folder}/*.hdr')
skybox_random_selection = skyboxes[random.randint(0,len(skyboxes)-1)]

dome_tex = visii.texture.create_from_file('dome_tex',skybox_random_selection)
visii.set_dome_light_texture(dome_tex)
visii.set_dome_light_intensity(random.uniform(1.1,2))
# visii.set_dome_light_intensity(1.15)
visii.set_dome_light_rotation(
    # visii.angleAxis(visii.pi()/2,visii.vec3(1,0,0)) \
    # * visii.angleAxis(visii.pi()/2,visii.vec3(0,0,1))\
    visii.angleAxis(random.uniform(-visii.pi(),visii.pi()),visii.vec3(0,0,1))\
    # * visii.angleAxis(visii.pi()/2,visii.vec3(0,1,0))\
    # * visii.angleAxis(random.uniform(-visii.pi()/8,visii.pi()/8),visii.vec3(0,0,1))\
)

# # # # # # # # # # # # # # # # # # # # # # # # #
# Lets set some objects in the scene

mesh_loaded = {}
objects_to_move = []

sample_space= [
        [-20,20],
        [-20,20],
        [-30,-2]
    ]

if opt.interactive:
    physicsClient = p.connect(p.GUI) # non-graphical version
else:
    physicsClient = p.connect(p.DIRECT) # non-graphical version

visii_pybullet = []
names_to_export = []

def adding_mesh_object(
        name, 
        obj_to_load, 
        texture_to_load, 
        model_info_path=None, 
        scale=1, 
        debug=False
    ):
    global mesh_loaded, visii_pybullet, names_to_export
    # obj_to_load = toy_to_load + "/meshes/model.obj"
    # texture_to_load = toy_to_load + "/materials/textures/texture.png"

    if texture_to_load is None:
        toys = load_obj_scene(obj_to_load)
        if len(toys) > 1: 
            print("more than one model in the object, \
                   materials might be wrong!")
        toy_transform = visii.entity.get(toys[0]).get_transform()
        toy_material = visii.entity.get(toys[0]).get_material()
        toy_mesh = visii.entity.get(toys[0]).get_mesh()        

        obj_export = visii.entity.create(
                name = name,
                transform = visii.transform.create(
                    name = name, 
                    position = toy_transform.get_position(),
                    rotation = toy_transform.get_rotation(),
                    scale = toy_transform.get_scale(),
                ),
                material = toy_material,
                mesh = visii.mesh.create_from_file(name,obj_to_load),
            )

        toy_transform = obj_export.get_transform()
        obj_export.get_material().set_roughness(random.uniform(0.1, 0.5))

        for toy in toys:
            visii.entity.remove(toy)

        toys = [name]
    else:
        toys = [name]

        if obj_to_load in mesh_loaded:
            toy_mesh = mesh_loaded[obj_to_load]
        else:
            toy_mesh = visii.mesh.create_from_file(name, obj_to_load)
            mesh_loaded[obj_to_load] = toy_mesh

        toy = visii.entity.create(
            name=name,
            transform=visii.transform.create(name),
            mesh=toy_mesh,
            material=visii.material.create(name)
        )

        toy_rgb_tex = visii.texture.create_from_file(name, texture_to_load)
        toy.get_material().set_base_color_texture(toy_rgb_tex)
        toy.get_material().set_roughness(random.uniform(0.1, 0.5))

        toy_transform = toy.get_transform()

    ###########################

    toy_transform.set_scale(visii.vec3(scale))
    toy_transform.set_position(
        visii.vec3(
            random.uniform(0.1, 2),
            random.uniform(-1, 1),
            random.uniform(-1, 1),
        )
    )
    toy_transform.set_rotation(
        visii.quat(
            random.uniform(0, 1),
            random.uniform(0, 1),
            random.uniform(0, 1),
            random.uniform(0, 1),
        )
    )

    # add symmetry_corrected transform
    child_transform = visii.transform.create(f"{toy_transform.get_name()}_symmetry_corrected")
    # print(toy_transform.get_name())
    # print(child_transform)
    child_transform.set_parent(toy_transform)

    # store symmetry transforms for later use.
    symmetry_transforms = get_symmetry_transformations(model_info_path)

    # create physics for object
    id_pybullet = create_physics(name, mass=(np.random.rand() * 5))

    if model_info_path is not None:
        try:
            with open(model_info_path) as json_file:
                model_info = json.load(json_file)
        except FileNotFoundError:
            model_info = {}
    else:
        model_info = {}

    visii_pybullet.append(
        {
            'visii_id': name,
            'bullet_id': id_pybullet,
            'base_rot': None,
            'model_info': model_info,
            'symmetry_transforms': symmetry_transforms
        }
    )
    gemPos, gemOrn = p.getBasePositionAndOrientation(id_pybullet)
    force_rand = 10
    object_position = 0.01
    p.applyExternalForce(
        id_pybullet,
        -1,
        [random.uniform(-force_rand, force_rand),
         random.uniform(-force_rand, force_rand),
         random.uniform(-force_rand, force_rand)],
        [random.uniform(-object_position, object_position),
         random.uniform(-object_position, object_position),
         random.uniform(-object_position, object_position)],
        flags=p.WORLD_FRAME
    )

    for entity_name in toys:
        names_to_export.append(entity_name)
        add_cuboid(entity_name, scale=scale, debug=debug)

google_content_folder = glob.glob(opt.objs_folder_distrators + "*/")

for i_obj in range(int(opt.nb_distractors)):

    toy_to_load = google_content_folder[random.randint(0,len(google_content_folder)-1)]

    obj_to_load = toy_to_load + "/meshes/model.obj"
    texture_to_load = toy_to_load + "/materials/textures/texture.png"
    name = "google_"+toy_to_load.split('/')[-2] + f"_{i_obj}"

    adding_mesh_object(name, obj_to_load, texture_to_load, debug=opt.debug)

if opt.path_single_obj is not None:
    for i_object in range(opt.nb_objects):
        model_info_path = os.path.dirname(opt.path_single_obj) + '/model_info.json'

        if not os.path.exists(model_info_path):
            model_info_path = None

        adding_mesh_object(f"single_obj_{i_object}",
                           opt.path_single_obj,
                           None,
                           model_info_path,
                           scale=opt.scale,
                           debug=opt.debug)
else:
    google_content_folder = glob.glob(opt.objs_folder + "*/")

    for i_obj in range(int(opt.nb_objects)):
        toy_to_load = google_content_folder[random.randint(0, len(google_content_folder) - 1)]

        obj_to_load = toy_to_load + "/google_16k/woodbar.obj"
        texture_to_load = toy_to_load + "/google_16k/woodbar.png"
        model_info_path = toy_to_load + "/google_16k/woodbar_info.json"
        name = "hope_" + toy_to_load.split('/')[-2] + f"_{i_obj}"

        adding_mesh_object(name, obj_to_load, texture_to_load, model_info_path, scale=opt.scale, debug=opt.debug)

        # p.applyExternalTorque(id_pybullet,-1,
        #     [   random.uniform(-force_rand,force_rand),
        #         random.uniform(-force_rand,force_rand),
        #         random.uniform(-force_rand,force_rand)],
        #     [0,0,0],
        #     flags=p.WORLD_FRAME
        # )

camera_pybullet_col = p.createCollisionShape(p.GEOM_SPHERE,0.05)
camera_pybullet = p.createMultiBody(
    baseCollisionShapeIndex = camera_pybullet_col,
    basePosition = camera.get_transform().get_position(),
    # baseOrientation= rot,
)

# print('simulate')
# for i in range (10000):
#     p.stepSimulation()
#     # time.sleep(0.1)
#     time.sleep(1./240.)

box_position = opt.box_size

plane1 = p.createCollisionShape(p.GEOM_PLANE,planeNormal = [0,0,-1])
rot1 = visii.angleAxis(-visii.pi()/16,visii.vec3(0,1,0))
plane1_body = p.createMultiBody(
    baseCollisionShapeIndex = plane1,
    basePosition = [0,0,-box_position],
    baseOrientation= [rot1[3],rot1[0],rot1[1],rot1[2]],
)

plane1 = p.createCollisionShape(p.GEOM_PLANE,planeNormal = [0,0,1])
rot1 = visii.angleAxis(visii.pi()/16,visii.vec3(0,1,0))
plane2_body = p.createMultiBody(
    baseCollisionShapeIndex = plane1,
    basePosition = [0,0,box_position],
    baseOrientation= [rot1[3],rot1[0],rot1[1],rot1[2]],
)

plane1 = p.createCollisionShape(p.GEOM_PLANE,planeNormal = [0,1,0])
rot1 = visii.angleAxis(visii.pi()/16,visii.vec3(1,0,0))
plane3_body = p.createMultiBody(
    baseCollisionShapeIndex = plane1,
    basePosition = [0,+box_position,0],
    baseOrientation= [rot1[3],rot1[0],rot1[1],rot1[2]],
)

plane1 = p.createCollisionShape(p.GEOM_PLANE,planeNormal = [0,-1,0])
rot1 = visii.angleAxis(-visii.pi()/16,visii.vec3(1,0,0))
plane4_body = p.createMultiBody(
    baseCollisionShapeIndex = plane1,
    basePosition = [0,-box_position,0],
    baseOrientation= [rot1[3],rot1[0],rot1[1],rot1[2]],
)

plane1 = p.createCollisionShape(p.GEOM_PLANE,planeNormal = [-1,0,0])
plane5_body = p.createMultiBody(
    baseCollisionShapeIndex = plane1,
    basePosition = [2,0,0],
    # baseOrientation= [rot1[3],rot1[0],rot1[1],rot1[2]],
)

plane1 = p.createCollisionShape(p.GEOM_PLANE,planeNormal = [1,0,0])
plane6_body = p.createMultiBody(
    baseCollisionShapeIndex = plane1,
    basePosition = [0.1,0,0],
    # baseOrientation= [rot1[3],rot1[0],rot1[1],rot1[2]],
)

# # # # # # # # # # # # # # # # # # # # # # # # #

export_to_ndds_folder_settings_files(
    opt.outf,
    obj_names=names_to_export,
    width=opt.width,
    height=opt.height,
    camera_name='camera',
)

i_frame = -1
i_render = 0

while True:
    p.stepSimulation()

    i_frame += 1

    if i_render >= int(opt.nb_frames) and not opt.interactive:
        break

    # update the position from pybullet
    for i_entry, entry in enumerate(visii_pybullet):
        # print('update',entry)
        visii.entity.get(entry['visii_id']).get_transform().set_position(
            visii.entity.get(entry['visii_id']).get_transform().get_position(),
            previous = True
        )
        visii.entity.get(entry['visii_id']).get_transform().set_rotation(
            visii.entity.get(entry['visii_id']).get_transform().get_rotation(),
            previous = True
        )

        update_pose(entry)
        if random.random() > 0.99:
            force_rand = 10
            object_position = 0.01
            p.applyExternalForce(
                entry['bullet_id'],
                -1,
                [   random.uniform(-force_rand,force_rand),
                    random.uniform(-force_rand,force_rand),
                    random.uniform(-force_rand,force_rand)],
                [   random.uniform(-object_position,object_position),
                    random.uniform(-object_position,object_position),
                    random.uniform(-object_position,object_position)],
                flags=p.WORLD_FRAME
            )
            # break
    if i_frame == 0:
        continue

    if not opt.interactive:
        if not i_frame % int(opt.skip_frame) == 0:
            continue

        print(f"{str(i_render).zfill(5)}/{str(opt.nb_frames).zfill(5)}")

        visii.sample_pixel_area(
            x_sample_interval = (0,1),
            y_sample_interval = (0,1))

        visii.render_to_file(
            width=int(opt.width),
            height=int(opt.height),
            samples_per_pixel=int(opt.spp),
            file_path=f"{opt.outf}/{str(i_render).zfill(5)}.png"
        )
        visii.sample_pixel_area(
            x_sample_interval = (.5,.5),
            y_sample_interval = (.5,.5))

        visii.sample_time_interval((1,1))

        visii.render_data_to_file(
            width=opt.width,
            height=opt.height,
            start_frame=0,
            frame_count=1,
            bounce=int(0),
            options="entity_id",
            file_path = f"{opt.outf}/{str(i_render).zfill(5)}.seg.exr"
        )
        segmentation_mask = visii.render_data(
            width=int(opt.width),
            height=int(opt.height),
            start_frame=0,
            frame_count=1,
            bounce=int(0),
            options="entity_id",
        )
        segmentation_mask = np.array(segmentation_mask).reshape((opt.height, opt.width, 4))[:, :, 0]
        export_to_ndds_file(
            f"{opt.outf}/{str(i_render).zfill(5)}.json",
            obj_names = names_to_export,
            width = opt.width,
            height = opt.height,
            camera_name = 'camera',
            # cuboids = cuboids,
            camera_struct = random_camera_movement,
            segmentation_mask=segmentation_mask,
            compute_visibility_fraction=opt.visibility_fraction,
        )
        visii.render_data_to_file(
            width=opt.width,
            height=opt.height,
            start_frame=0,
            frame_count=1,
            bounce=int(0),
            options="depth",
            file_path = f"{opt.outf}/{str(i_render).zfill(5)}.depth.exr"
        )

        i_render +=1

# subprocess.call(['ffmpeg', '-y',\
#     '-framerate', '30', "-hide_banner", "-loglevel", \
#     "panic",'-pattern_type', 'glob', '-i',\
#     f"{opt.outf}/*.png", f"{opt.outf}/video.mp4"])

visii.deinitialize()

and my generate_dataset.py file looks like this:

#!/usr/bin/env python3
import random 
import subprocess

# 20 000 images

for i in range(0, 100):
    to_call = [
        "python",'single_video_pybullet.py',
        '--spp','10',
        '--nb_frames', '200',
        '--nb_objects',str(int(random.uniform(50,75))),
        '--scale', '0.01',
        '--outf',f"dataset/{str(i).zfill(3)}",
    ]
    subprocess.call(to_call)

I am stuck into this for quite sometime. Can you tell me how can I fix it ?

TontonTremblay commented 1 year ago

Which version of python are you using?

ArghyaChatterjee commented 1 year ago

@TontonTremblay python3.

TontonTremblay commented 1 year ago

It looks like you have a problem with f string, which are not supported in python2

TontonTremblay commented 1 year ago

If you do not run the generate script, does it work?

ArghyaChatterjee commented 1 year ago

@TontonTremblay , yes it works. I think I made my default python as python 2 for some other projects which affected this one. So, I made it back to python3 when calling python. Thanks.

BTW, inside generate_dataset.py, what does this line mean:

'--nb_objects',str(int(random.uniform(50,75))),

This was giving some trouble. My model was a bit big so probably it was getting memory error to randomize bigger models with bigger numbers.

TontonTremblay commented 1 year ago

Yeah these values are pretty high. it samples an integer between 50 and 75 and will load that many 3d models as distractors. This will take a while to load.