id
int64
0
190k
prompt
stringlengths
21
13.4M
docstring
stringlengths
1
12k
22,926
from . import * def build_scenario(builder): builder.config().game_duration = 3000 builder.config().right_team_difficulty = 0.95 builder.config().deterministic = False if builder.EpisodeNumber() % 2 == 0: first_team = Team.e_Left second_team = Team.e_Right else: first_team = Team.e_Right second_team = Team.e_Left builder.SetTeam(first_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(0.000000, 0.020000, e_PlayerRole_RM) builder.AddPlayer(0.000000, -0.020000, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM) builder.SetTeam(second_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(-0.050000, 0.000000, e_PlayerRole_RM) builder.AddPlayer(-0.010000, 0.216102, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM)
null
22,927
from . import * def build_scenario(builder): builder.config().game_duration = 3000 builder.config().right_team_difficulty = 1.0 builder.config().left_team_difficulty = 1.0 builder.config().deterministic = False if builder.EpisodeNumber() % 2 == 0: first_team = Team.e_Left second_team = Team.e_Right else: first_team = Team.e_Right second_team = Team.e_Left builder.SetTeam(first_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(0.000000, 0.020000, e_PlayerRole_RM) builder.AddPlayer(0.000000, -0.020000, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM) builder.SetTeam(second_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(-0.050000, 0.000000, e_PlayerRole_RM) builder.AddPlayer(-0.010000, 0.216102, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM)
null
22,928
from . import * def build_scenario(builder): builder.config().game_duration = 3000 builder.config().right_team_difficulty = 0.05 builder.config().left_team_difficulty = 0.05 builder.config().deterministic = False if builder.EpisodeNumber() % 2 == 0: first_team = Team.e_Left second_team = Team.e_Right else: first_team = Team.e_Right second_team = Team.e_Left builder.SetTeam(first_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK, controllable=False) builder.AddPlayer(0.000000, 0.020000, e_PlayerRole_RM) builder.AddPlayer(0.000000, -0.020000, e_PlayerRole_CF) builder.AddPlayer(-0.1, -0.1, e_PlayerRole_LB) builder.AddPlayer(-0.1, 0.1, e_PlayerRole_CB) builder.SetTeam(second_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK, controllable=False) builder.AddPlayer(-0.04, 0.040000, e_PlayerRole_RM) builder.AddPlayer(-0.04, -0.040000, e_PlayerRole_CF) builder.AddPlayer(-0.1, -0.1, e_PlayerRole_LB) builder.AddPlayer(-0.1, 0.1, e_PlayerRole_CB)
null
22,929
from . import * def build_scenario(builder): builder.config().game_duration = 400 builder.config().deterministic = False builder.config().offsides = False builder.config().end_episode_on_score = True builder.config().end_episode_on_out_of_play = True builder.config().end_episode_on_possession_change = True builder.SetBallPosition(0.02, 0.0) builder.SetTeam(Team.e_Left) builder.AddPlayer(-1.0, 0.0, e_PlayerRole_GK) builder.AddPlayer(0.0, 0.0, e_PlayerRole_CB) builder.SetTeam(Team.e_Right) builder.AddPlayer(-1.0, 0.0, e_PlayerRole_GK) builder.AddPlayer(0.12, 0.2, e_PlayerRole_LB) builder.AddPlayer(0.12, 0.1, e_PlayerRole_CB) builder.AddPlayer(0.12, 0.0, e_PlayerRole_CM) builder.AddPlayer(0.12, -0.1, e_PlayerRole_CB) builder.AddPlayer(0.12, -0.2, e_PlayerRole_RB)
null
22,930
from . import * episode = 0 def build_scenario(builder): global episode episode += 1 builder.config().game_duration = 3000 builder.config().deterministic = False builder.config().offsides = False builder.config().end_episode_on_score = True builder.config().end_episode_on_out_of_play = True builder.config().end_episode_on_possession_change = True builder.SetBallPosition(-0.48, -0.06356) builder.SetTeam(Team.e_Left) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(0.000000, 0.020000, e_PlayerRole_RM) builder.AddPlayer(0.000000, -0.020000, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM) # All right players are lazy (i.e., they don't move, except the keeper) builder.SetTeam(Team.e_Right) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(-0.050000, 0.000000, e_PlayerRole_RM, True) builder.AddPlayer(-0.010000, 0.216102, e_PlayerRole_CF, True) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB, True) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB, True) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB, True) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB, True) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM, True) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM, True) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM, True) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM, True)
null
22,931
from . import * def build_scenario(builder): builder.config().game_duration = 3000 builder.config().right_team_difficulty = 0.6 builder.config().deterministic = False if builder.EpisodeNumber() % 2 == 0: first_team = Team.e_Left second_team = Team.e_Right else: first_team = Team.e_Right second_team = Team.e_Left builder.SetTeam(first_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(0.000000, 0.020000, e_PlayerRole_RM) builder.AddPlayer(0.000000, -0.020000, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM) builder.SetTeam(second_team) builder.AddPlayer(-1.000000, 0.000000, e_PlayerRole_GK) builder.AddPlayer(-0.050000, 0.000000, e_PlayerRole_RM) builder.AddPlayer(-0.010000, 0.216102, e_PlayerRole_CF) builder.AddPlayer(-0.422000, -0.19576, e_PlayerRole_LB) builder.AddPlayer(-0.500000, -0.06356, e_PlayerRole_CB) builder.AddPlayer(-0.500000, 0.063559, e_PlayerRole_CB) builder.AddPlayer(-0.422000, 0.195760, e_PlayerRole_RB) builder.AddPlayer(-0.184212, -0.10568, e_PlayerRole_CM) builder.AddPlayer(-0.267574, 0.000000, e_PlayerRole_CM) builder.AddPlayer(-0.184212, 0.105680, e_PlayerRole_CM) builder.AddPlayer(-0.010000, -0.21610, e_PlayerRole_LM)
null
22,932
from . import * def build_scenario(builder): builder.config().game_duration = 400 builder.config().deterministic = False builder.config().offsides = False builder.config().end_episode_on_score = True builder.config().end_episode_on_out_of_play = True builder.config().end_episode_on_possession_change = True builder.SetBallPosition(0.7, -0.28) builder.SetTeam(Team.e_Left) builder.AddPlayer(-1.0, 0.0, e_PlayerRole_GK) builder.AddPlayer(0.7, 0.0, e_PlayerRole_CB) builder.AddPlayer(0.7, -0.3, e_PlayerRole_CB) builder.SetTeam(Team.e_Right) builder.AddPlayer(-1.0, 0.0, e_PlayerRole_GK) builder.AddPlayer(-0.75, 0.1, e_PlayerRole_CB)
null
22,933
from baselines.common.models import register import sonnet as snt import tensorflow.compat.v1 as tf def gfootball_impala_cnn(): def network_fn(frame): # Convert to floats. frame = tf.to_float(frame) frame /= 255 with tf.variable_scope('convnet'): conv_out = frame conv_layers = [(16, 2), (32, 2), (32, 2), (32, 2)] for i, (num_ch, num_blocks) in enumerate(conv_layers): # Downscale. conv_out = snt.Conv2D(num_ch, 3, stride=1, padding='SAME')(conv_out) conv_out = tf.nn.pool( conv_out, window_shape=[3, 3], pooling_type='MAX', padding='SAME', strides=[2, 2]) # Residual block(s). for j in range(num_blocks): with tf.variable_scope('residual_%d_%d' % (i, j)): block_input = conv_out conv_out = tf.nn.relu(conv_out) conv_out = snt.Conv2D(num_ch, 3, stride=1, padding='SAME')(conv_out) conv_out = tf.nn.relu(conv_out) conv_out = snt.Conv2D(num_ch, 3, stride=1, padding='SAME')(conv_out) conv_out += block_input conv_out = tf.nn.relu(conv_out) conv_out = snt.BatchFlatten()(conv_out) conv_out = snt.Linear(256)(conv_out) conv_out = tf.nn.relu(conv_out) return conv_out return network_fn
null
22,934
from __future__ import absolute_import from __future__ import division from __future__ import print_function import os import tempfile import argparse import gfootball.env as football_env import gym import ray from ray import tune from ray.rllib.env.multi_agent_env import MultiAgentEnv from ray.tune.registry import register_env def gen_policy(_): return (None, obs_space, act_space, {})
null
22,935
from __future__ import absolute_import from __future__ import division from __future__ import print_function import multiprocessing import os from absl import app from absl import flags from baselines import logger from baselines.bench import monitor from baselines.common.vec_env.subproc_vec_env import SubprocVecEnv from baselines.ppo2 import ppo2 import gfootball.env as football_env from gfootball.examples import models FLAGS = flags.FLAGS def create_single_football_env(iprocess): """Creates gfootball environment.""" env = football_env.create_environment( env_name=FLAGS.level, stacked=('stacked' in FLAGS.state), rewards=FLAGS.reward_experiment, logdir=logger.get_dir(), write_goal_dumps=FLAGS.dump_scores and (iprocess == 0), write_full_episode_dumps=FLAGS.dump_full_episodes and (iprocess == 0), render=FLAGS.render and (iprocess == 0), dump_frequency=50 if FLAGS.render and iprocess == 0 else 0) env = monitor.Monitor(env, logger.get_dir() and os.path.join(logger.get_dir(), str(iprocess))) return env The provided code snippet includes necessary dependencies for implementing the `train` function. Write a Python function `def train(_)` to solve the following problem: Trains a PPO2 policy. Here is the function: def train(_): """Trains a PPO2 policy.""" vec_env = SubprocVecEnv([ (lambda _i=i: create_single_football_env(_i)) for i in range(FLAGS.num_envs) ], context=None) # Import tensorflow after we create environments. TF is not fork sake, and # we could be using TF as part of environment if one of the players is # controlled by an already trained model. import tensorflow.compat.v1 as tf ncpu = multiprocessing.cpu_count() config = tf.ConfigProto(allow_soft_placement=True, intra_op_parallelism_threads=ncpu, inter_op_parallelism_threads=ncpu) config.gpu_options.allow_growth = True tf.Session(config=config).__enter__() ppo2.learn(network=FLAGS.policy, total_timesteps=FLAGS.num_timesteps, env=vec_env, seed=FLAGS.seed, nsteps=FLAGS.nsteps, nminibatches=FLAGS.nminibatches, noptepochs=FLAGS.noptepochs, max_grad_norm=FLAGS.max_grad_norm, gamma=FLAGS.gamma, ent_coef=FLAGS.ent_coef, lr=FLAGS.lr, log_interval=1, save_interval=FLAGS.save_interval, cliprange=FLAGS.cliprange, load_path=FLAGS.load_path)
Trains a PPO2 policy.
22,936
import random import string import time import urllib.request from gfootball.eval_server import config import grpc def get_grpc_channel(server): # send keepalive ping every 10 second # allow unlimited amount of keepalive pings options = (('grpc.keepalive_time_ms', 10000), ('grpc.http2.max_pings_without_data', 0)) return grpc.insecure_channel(server, options=options)
null
22,937
import random import string import time import urllib.request from gfootball.eval_server import config import grpc def get_random_string(length=10, append_timestamp=True): characters = string.ascii_lowercase + string.ascii_uppercase + string.digits res = ''.join(random.choice(characters) for i in range(length)) if append_timestamp: res += '_{}'.format(int(time.time())) return res def get_master_address(track='default'): # We add ''?' + get_random_string()' to avoid caching. Somehow even with # disabled server side caching on the file we were still getting the old # content of the file. response = urllib.request.urlopen( config.master_address_public_path + '_' + track + '?' + get_random_string()) ip = response.read().decode('utf-8').strip() return '{}:{}'.format(ip, config.grpc_port)
null
22,938
import grpc from gfootball.eval_server.proto import master_pb2 as gfootball_dot_eval__server_dot_proto_dot_master__pb2 def add_MasterServicer_to_server(servicer, server): rpc_method_handlers = { 'StartGame': grpc.unary_unary_rpc_method_handler( servicer.StartGame, request_deserializer=gfootball_dot_eval__server_dot_proto_dot_master__pb2.StartGameRequest.FromString, response_serializer=gfootball_dot_eval__server_dot_proto_dot_master__pb2.StartGameResponse.SerializeToString, ), } generic_handler = grpc.method_handlers_generic_handler( 'gfootball.eval_server.Master', rpc_method_handlers) server.add_generic_rpc_handlers((generic_handler,))
null
22,939
import grpc from gfootball.eval_server.proto import game_server_pb2 as gfootball_dot_eval__server_dot_proto_dot_game__server__pb2 def add_GameServerServicer_to_server(servicer, server): rpc_method_handlers = { 'GetEnvResult': grpc.unary_unary_rpc_method_handler( servicer.GetEnvResult, request_deserializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.GetEnvResultRequest.FromString, response_serializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.GetEnvResultResponse.SerializeToString, ), 'Step': grpc.unary_unary_rpc_method_handler( servicer.Step, request_deserializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.StepRequest.FromString, response_serializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.StepResponse.SerializeToString, ), 'GetCapacity': grpc.unary_unary_rpc_method_handler( servicer.GetCapacity, request_deserializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.GetCapacityRequest.FromString, response_serializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.GetCapacityResponse.SerializeToString, ), 'CreateGame': grpc.unary_unary_rpc_method_handler( servicer.CreateGame, request_deserializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.CreateGameRequest.FromString, response_serializer=gfootball_dot_eval__server_dot_proto_dot_game__server__pb2.CreateGameResponse.SerializeToString, ), } generic_handler = grpc.method_handlers_generic_handler( 'gfootball.eval_server.GameServer', rpc_method_handlers) server.add_generic_rpc_handlers((generic_handler,))
null
22,940
import random from absl import app from absl import flags from absl import logging import gfootball.env as football_env from gfootball.env import football_action_set import grpc import numpy as np import tensorflow.compat.v2 as tf FLAGS = flags.FLAGS def random_actions(obs): num_players = 1 if len(obs.shape) == 3 else obs.shape[0] a = [] for _ in range(num_players): a.append(random.randint(0, NUM_ACTIONS - 1)) return a def generate_actions(obs, model): a = [] # Single agent case if len(obs.shape) == 3: a.append(model(seed_rl_preprocessing(obs))[0][0].numpy()) else: # Multiagent -> first dimension is a number of agents you control. for x in range(obs.shape[0]): a.append(model(seed_rl_preprocessing(obs[x]))[0][0].numpy()) return a def get_inference_model(inference_model): if not inference_model or FLAGS.username == 'random': return random_actions model = tf.saved_model.load(inference_model) return lambda obs: generate_actions(obs, model)
null
22,941
from __future__ import absolute_import from __future__ import division from __future__ import print_function from gfootball_engine import e_BackendAction import numpy from six.moves import range action_left = CoreAction( e_BackendAction.left, "left", sticky=True, directional=True) action_release_direction = CoreAction( e_BackendAction.release_direction, "release_direction", directional=True) reverse_action_mapping = { action_long_pass: action_release_long_pass, action_high_pass: action_release_high_pass, action_short_pass: action_release_short_pass, action_shot: action_release_shot, action_keeper_rush: action_release_keeper_rush, action_sliding: action_release_sliding, action_pressure: action_release_pressure, action_team_pressure: action_release_team_pressure, action_switch: action_release_switch, action_sprint: action_release_sprint, action_dribble: action_release_dribble, action_release_long_pass: action_long_pass, action_release_high_pass: action_high_pass, action_release_short_pass: action_short_pass, action_release_shot: action_shot, action_release_keeper_rush: action_keeper_rush, action_release_sliding: action_sliding, action_release_pressure: action_pressure, action_release_team_pressure: action_team_pressure, action_release_switch: action_switch, action_release_sprint: action_sprint, action_release_dribble: action_dribble } def disable_action(action): assert set(action.__dict__) == set(action_left.__dict__) if action._directional: return action_release_direction return reverse_action_mapping[action]
null
22,942
from __future__ import absolute_import from __future__ import division from __future__ import print_function from gfootball.env import football_action_set import numpy as np def rotate_3d_point(point): """Rotate 3d point around the center of the field. Args: points: [x, y, z] point. Returns: The rotated points. """ # This assumes the center of the field is the origin: (0, 0) return np.array([-point[0], -point[1], point[2]]) def flip_team_observation(observation, result, config, from_team, to_team): """Rotates team-specific observations.""" result['{}_team'.format(to_team)] = rotate_points( observation['{}_team'.format(from_team)]) result['{}_team_direction'.format(to_team)] = rotate_points( observation['{}_team_direction'.format(from_team)]) result['{}_team_tired_factor'.format(to_team)] = observation[ '{}_team_tired_factor'.format(from_team)] result['{}_team_active'.format(to_team)] = observation[ '{}_team_active'.format(from_team)] result['{}_team_yellow_card'.format(to_team)] = observation[ '{}_team_yellow_card'.format(from_team)] result['{}_team_roles'.format(to_team)] = observation['{}_team_roles'.format( from_team)] result['{}_team_active'.format(to_team)] = observation[ '{}_team_active'.format(from_team)] result['{}_team_designated_player'.format(to_team)] = observation[ '{}_team_designated_player'.format(from_team)] if '{}_agent_controlled_player'.format(from_team) in observation: result['{}_agent_controlled_player'.format(to_team)] = observation[ '{}_agent_controlled_player'.format(from_team)] if '{}_agent_sticky_actions'.format(from_team) in observation: result['{}_agent_sticky_actions'.format(to_team)] = [ rotate_sticky_actions(sticky, config) for sticky in observation['{}_agent_sticky_actions'.format(from_team)] ] The provided code snippet includes necessary dependencies for implementing the `flip_observation` function. Write a Python function `def flip_observation(observation, config)` to solve the following problem: Observation corresponding to the field rotated by 180 degrees. Here is the function: def flip_observation(observation, config): """Observation corresponding to the field rotated by 180 degrees.""" flipped_observation = {} flipped_observation['ball'] = rotate_3d_point(observation['ball']) flipped_observation['ball_direction'] = rotate_3d_point( observation['ball_direction']) flipped_observation['ball_rotation'] = observation['ball_rotation'] flipped_observation['ball_owned_team'] = 1 - observation[ 'ball_owned_team'] if observation['ball_owned_team'] > -1 else -1 flipped_observation['ball_owned_player'] = observation['ball_owned_player'] flipped_observation['score'] = [ observation['score'][1], observation['score'][0] ] flipped_observation['game_mode'] = observation['game_mode'] flipped_observation['steps_left'] = observation['steps_left'] flip_team_observation(observation, flipped_observation, config, 'left', 'right') flip_team_observation(observation, flipped_observation, config, 'right', 'left') return flipped_observation
Observation corresponding to the field rotated by 180 degrees.
22,943
from __future__ import absolute_import from __future__ import division from __future__ import print_function from gfootball.env import football_action_set import numpy as np def flip_single_action(action, config): def flip_action(action, config): if isinstance(action, np.ndarray) or isinstance(action, list): return [flip_single_action(a, config) for a in action] return flip_single_action(action, config)
null
22,944
import pygame _controllers = [] def add_controller(controller_kind, controller_index=None): global _controllers _controllers.append((controller_kind, controller_index))
null
22,945
import pygame _queue = [] _controllers = [] def fits(event, controller_kind, controller_index): if controller_kind == 'keyboard': return event.type in KEYBOARD_EVENTS if controller_kind == 'gamepad': return event.type in GAMEPAD_EVENTS and event.joy == controller_index assert False, 'Unknown controller kind!' def get(controller_kind, controller_index=None): global _queue global _controllers _queue.extend(pygame.event.get()) result = [] new_state = [] for event in _queue: if fits(event, controller_kind, controller_index): result.append(event) else: for controller in _controllers: if fits(event, *controller): new_state.append(event) break _queue = new_state return result
null
22,946
from __future__ import print_function import copy import tempfile import os import platform from absl import flags import gfootball_engine as libgame def parse_player_definition(definition): """Parses player definition. An example of player definition is: "agent:players=4" or "replay:path=...". Args: definition: a string defining a player Returns: A tuple (name, dict). """ name = definition d = {'left_players': 0, 'right_players': 0} if ':' in definition: # Windows requires special handling of replays, because path may contain ':' if platform.system() == 'Windows' and definition.startswith('replay:') \ and len(definition.split(':')) > 2: (name, params) = 'replay', definition.split('replay:')[-1] else: (name, params) = definition.split(':') for param in params.split(','): (key, value) = param.split('=') d[key] = value if d['left_players'] == 0 and d['right_players'] == 0: d['left_players'] = 1 return name, d The provided code snippet includes necessary dependencies for implementing the `count_left_players` function. Write a Python function `def count_left_players(definition)` to solve the following problem: Returns a number of left players given a definition. Here is the function: def count_left_players(definition): """Returns a number of left players given a definition.""" return int(parse_player_definition(definition)[1]['left_players'])
Returns a number of left players given a definition.
22,947
from __future__ import print_function import copy import tempfile import os import platform from absl import flags import gfootball_engine as libgame def parse_player_definition(definition): """Parses player definition. An example of player definition is: "agent:players=4" or "replay:path=...". Args: definition: a string defining a player Returns: A tuple (name, dict). """ name = definition d = {'left_players': 0, 'right_players': 0} if ':' in definition: # Windows requires special handling of replays, because path may contain ':' if platform.system() == 'Windows' and definition.startswith('replay:') \ and len(definition.split(':')) > 2: (name, params) = 'replay', definition.split('replay:')[-1] else: (name, params) = definition.split(':') for param in params.split(','): (key, value) = param.split('=') d[key] = value if d['left_players'] == 0 and d['right_players'] == 0: d['left_players'] = 1 return name, d The provided code snippet includes necessary dependencies for implementing the `count_right_players` function. Write a Python function `def count_right_players(definition)` to solve the following problem: Returns a number of players given a definition. Here is the function: def count_right_players(definition): """Returns a number of players given a definition.""" return int(parse_player_definition(definition)[1]['right_players'])
Returns a number of players given a definition.
22,948
from __future__ import print_function import copy import tempfile import os import platform from absl import flags import gfootball_engine as libgame def count_players(definition): """Returns a number of players given a definition.""" _, player_definition = parse_player_definition(definition) return (int(player_definition['left_players']) + int(player_definition['right_players'])) The provided code snippet includes necessary dependencies for implementing the `get_agent_number_of_players` function. Write a Python function `def get_agent_number_of_players(players)` to solve the following problem: Returns a total number of players controlled by an agent. Here is the function: def get_agent_number_of_players(players): """Returns a total number of players controlled by an agent.""" return sum([count_players(player) for player in players if player.startswith('agent')])
Returns a total number of players controlled by an agent.
22,949
from __future__ import absolute_import from __future__ import division from __future__ import print_function from gfootball.env import football_action_set import numpy as np from six.moves import range SMM_WIDTH = 96 SMM_HEIGHT = 72 def get_smm_layers(config): return SMM_LAYERS def mark_points(frame, points): """Draw dots corresponding to 'points'. Args: frame: 2-d matrix representing one SMM channel ([y, x]) points: a list of (x, y) coordinates to be marked """ for p in range(len(points) // 2): x = int((points[p * 2] - MINIMAP_NORM_X_MIN) / (MINIMAP_NORM_X_MAX - MINIMAP_NORM_X_MIN) * frame.shape[1]) y = int((points[p * 2 + 1] - MINIMAP_NORM_Y_MIN) / (MINIMAP_NORM_Y_MAX - MINIMAP_NORM_Y_MIN) * frame.shape[0]) x = max(0, min(frame.shape[1] - 1, x)) y = max(0, min(frame.shape[0] - 1, y)) frame[y, x] = _MARKER_VALUE The provided code snippet includes necessary dependencies for implementing the `generate_smm` function. Write a Python function `def generate_smm(observation, config=None, channel_dimensions=(SMM_WIDTH, SMM_HEIGHT))` to solve the following problem: Returns a list of minimap observations given the raw features for each active player. Args: observation: raw features from the environment config: environment config channel_dimensions: resolution of SMM to generate Returns: (N, H, W, C) - shaped np array representing SMM. N stands for the number of players we are controlling. Here is the function: def generate_smm(observation, config=None, channel_dimensions=(SMM_WIDTH, SMM_HEIGHT)): """Returns a list of minimap observations given the raw features for each active player. Args: observation: raw features from the environment config: environment config channel_dimensions: resolution of SMM to generate Returns: (N, H, W, C) - shaped np array representing SMM. N stands for the number of players we are controlling. """ frame = np.zeros((len(observation), channel_dimensions[1], channel_dimensions[0], len(get_smm_layers(config))), dtype=np.uint8) for o_i, o in enumerate(observation): for index, layer in enumerate(get_smm_layers(config)): assert layer in o if layer == 'active': if o[layer] == -1: continue mark_points(frame[o_i, :, :, index], np.array(o['left_team'][o[layer]]).reshape(-1)) else: mark_points(frame[o_i, :, :, index], np.array(o[layer]).reshape(-1)) return frame
Returns a list of minimap observations given the raw features for each active player. Args: observation: raw features from the environment config: environment config channel_dimensions: resolution of SMM to generate Returns: (N, H, W, C) - shaped np array representing SMM. N stands for the number of players we are controlling.
22,950
from __future__ import absolute_import from __future__ import division from __future__ import print_function import collections import datetime import os import shutil import tempfile import timeit import traceback from absl import logging from gfootball.env import constants as const from gfootball.env import football_action_set from gfootball.scenarios import e_PlayerRole_GK import numpy as np from six.moves import range from six.moves import zip import six.moves.cPickle def write_players_state(writer, players_info): table_text = [["PLAYER", "SPRINT", "DRIBBLE", "DIRECTION", "ACTION"]] widths = [65, 65, 70, 85, 85] # Sort the players according to the order they appear in observations for _, player_info in sorted(players_info.items()): table_text.append([ (player_info['id'], player_info['color']), str(player_info.get("sprint", "-")), str(player_info.get("dribble", "-")), player_info.get("DIRECTION", "O"), player_info.get("ACTION", "-")]) writer.write_table(table_text, widths, scale_factor=1.0, offset=10)
null
22,951
from __future__ import absolute_import from __future__ import division from __future__ import print_function import collections import datetime import os import shutil import tempfile import timeit import traceback from absl import logging from gfootball.env import constants as const from gfootball.env import football_action_set from gfootball.scenarios import e_PlayerRole_GK import numpy as np from six.moves import range from six.moves import zip import six.moves.cPickle try: import cv2 except ImportError: import cv2 class TextWriter(object): def __init__(self, frame, x, y=0, field_coords=False, color=(255, 255, 255)): self._frame = frame if field_coords: x = 400 * (x + 1) - 10 y = 695 * (y + 0.43) self._pos_x = int(x) self._pos_y = int(y) + 20 self._color = color self._font = cv2.FONT_HERSHEY_SIMPLEX self._lineType = 2 self._arrow_types = ('top', 'top_right', 'right', 'bottom_right', 'bottom', 'bottom_left', 'left', 'top_left') def write(self, text, scale_factor=1, color=None): textPos = (self._pos_x, self._pos_y) fontScale = 0.8 * scale_factor cv2.putText(self._frame, text, textPos, self._font, fontScale, color or self._color, self._lineType) self._pos_y += int(25 * scale_factor) def write_table(self, data, widths, scale_factor=1, offset=0): # data is a list of rows. Each row is a list of strings. fontScale = 0.5 * scale_factor init_x = self._pos_x for row in data: assert (len(row) == len(widths)) self._pos_x += offset for col, cell in enumerate(row): color = self._color if isinstance(cell, tuple): assert (len(cell) == 2) (text, color) = cell else: assert (isinstance(cell, str)) text = cell if text in self._arrow_types: self.write_arrow(text, scale_factor=scale_factor) else: textPos = (self._pos_x, self._pos_y) cv2.putText(self._frame, text, textPos, self._font, fontScale, color, self._lineType) self._pos_x += widths[col] self._pos_x = init_x self._pos_y += int(20 * scale_factor) self._pos_x = init_x def write_arrow(self, arrow_type, scale_factor=1): assert (arrow_type in self._arrow_types) thickness = 1 arrow_offsets = { 'top': (12, 0, 12, -16), 'top_right': (4, -4, 16, -16), 'right': (0, -10, 20, -10), 'bottom_right': (4, -16, 16, -4), 'bottom': (10, -16, 10, 0), 'bottom_left': (12, -12, 0, 0), 'left': (20, -10, 0, -10), 'top_left': (16, -4, 4, -16) } (s_x, s_y, e_x, e_y) = tuple(int(v * scale_factor) for v in arrow_offsets[arrow_type]) start_point = (self._pos_x + s_x, self._pos_y + s_y) end_point = (self._pos_x + e_x, self._pos_y + e_y) image = cv2.arrowedLine(self._frame, start_point, end_point, self._color, thickness) e_PlayerRole_GK = libgame.e_PlayerRole.e_PlayerRole_GK def get_frame(trace): if 'frame' in trace._trace['observation']: return trace._trace['observation']['frame'] frame = np.uint8(np.zeros((600, 800, 3))) corner1 = (0, 0) corner2 = (799, 0) corner3 = (799, 599) corner4 = (0, 599) line_color = (0, 255, 255) cv2.line(frame, corner1, corner2, line_color) cv2.line(frame, corner2, corner3, line_color) cv2.line(frame, corner3, corner4, line_color) cv2.line(frame, corner4, corner1, line_color) cv2.line(frame, (399, 0), (399, 799), line_color) writer = TextWriter( frame, trace['ball'][0], trace['ball'][1], field_coords=True, color=(248, 244, 236)) writer.write('B') for player_idx, player_coord in enumerate(trace['left_team']): writer = TextWriter( frame, player_coord[0], player_coord[1], field_coords=True, color=(238, 68, 47)) letter = str(player_idx) if trace['left_team_roles'][player_idx] == e_PlayerRole_GK: letter = 'G' writer.write(letter) for player_idx, player_coord in enumerate(trace['right_team']): writer = TextWriter( frame, player_coord[0], player_coord[1], field_coords=True, color=(99, 172, 190)) letter = str(player_idx) if trace['right_team_roles'][player_idx] == e_PlayerRole_GK: letter = 'G' writer.write(letter) return frame
null
22,952
from __future__ import absolute_import from __future__ import division from __future__ import print_function import collections import datetime import os import shutil import tempfile import timeit import traceback from absl import logging from gfootball.env import constants as const from gfootball.env import football_action_set from gfootball.scenarios import e_PlayerRole_GK import numpy as np from six.moves import range from six.moves import zip import six.moves.cPickle def softmax(x): return np.exp(x) / np.sum(np.exp(x), axis=0)
null
22,953
import importlib import os import pkgutil import random import sys from absl import flags from absl import logging import gfootball_engine as libgame def all_scenarios(): path = os.path.abspath(__file__) path = os.path.join(os.path.dirname(os.path.dirname(path)), 'scenarios') scenarios = [] for m in pkgutil.iter_modules([path]): # There was API change in pkgutil between Python 3.5 and 3.6... if m.__class__ == tuple: scenarios.append(m[1]) else: scenarios.append(m.name) return scenarios
null
22,954
from baselines.common.policies import build_policy from gfootball.env import football_action_set from gfootball.env import observation_preprocessing from gfootball.env import player_base from gfootball.examples import models import gym import joblib import numpy as np import tensorflow.compat.v1 as tf The provided code snippet includes necessary dependencies for implementing the `_load_variables` function. Write a Python function `def _load_variables(load_path, sess, prefix='', remove_prefix=True)` to solve the following problem: Loads variables from checkpoint of policy trained by baselines. Here is the function: def _load_variables(load_path, sess, prefix='', remove_prefix=True): """Loads variables from checkpoint of policy trained by baselines.""" # Forked from address below since we needed loading from different var names: # https://github.com/openai/baselines/blob/master/baselines/common/tf_util.py variables = [v for v in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES) if v.name.startswith(prefix)] loaded_params = joblib.load(load_path) restores = [] for v in variables: v_name = v.name[len(prefix):] if remove_prefix else v.name restores.append(v.assign(loaded_params[v_name])) sess.run(restores)
Loads variables from checkpoint of policy trained by baselines.
22,955
import torch from thop import profile import torchvision import models import argparse def clever_format(nums, format="%.2f"): clever_nums = [] for num in nums: if num > 1e12: clever_nums.append(format % (num / 1024 ** 4) + "T") elif num > 1e9: clever_nums.append(format % (num / 1024 ** 3) + "G") elif num > 1e6: clever_nums.append(format % (num / 1024 ** 2) + "M") elif num > 1e3: clever_nums.append(format % (num / 1024) + "K") else: clever_nums.append(format % num + "B") clever_nums = clever_nums[0] if len(clever_nums) == 1 else (*clever_nums, ) return clever_nums
null
22,956
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import models class AverageMeter(object): def __init__(self): def reset(self): def update(self, val, n=1): def accuracy(output, target, topk=(1,)): def train(train_loader, model, criterion, optimizer, epoch): batch_time = AverageMeter() data_time = AverageMeter() losses = AverageMeter() top1 = AverageMeter() top5 = AverageMeter() losses_batch = {} # switch to train mode model.train() end = time.time() for i, (input, target) in enumerate(train_loader): # measure data loading time data_time.update(time.time() - end) if args.gpu is not None: input = input.cuda(args.gpu, non_blocking=True) target = target.cuda(args.gpu, non_blocking=True) # compute output output = model(input) loss = criterion(output, target) # measure accuracy and record loss prec1, prec5 = accuracy(output, target, topk=(1, 5)) losses.update(loss.item(), input.size(0)) top1.update(prec1[0], input.size(0)) top5.update(prec5[0], input.size(0)) # compute gradient and do SGD step optimizer.zero_grad() loss.backward() optimizer.step() # measure elapsed time batch_time.update(time.time() - end) end = time.time() if i % args.print_freq == 0: print('Epoch: [{0}][{1}/{2}]\t' 'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t' 'Data {data_time.val:.3f} ({data_time.avg:.3f})\t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format( epoch, i, len(train_loader), batch_time=batch_time, data_time=data_time, loss=losses, top1=top1, top5=top5)) return losses.avg, top1.avg, top5.avg
null
22,957
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import models class AverageMeter(object): """Computes and stores the average and current value""" def __init__(self): self.reset() def reset(self): self.val = 0 self.avg = 0 self.sum = 0 self.count = 0 def update(self, val, n=1): self.val = val self.sum += val * n self.count += n self.avg = self.sum / self.count def accuracy(output, target, topk=(1,)): """Computes the precision@k for the specified values of k""" with torch.no_grad(): maxk = max(topk) batch_size = target.size(0) _, pred = output.topk(maxk, 1, True, True) pred = pred.t() correct = pred.eq(target.view(1, -1).expand_as(pred)) res = [] for k in topk: correct_k = correct[:k].view(-1).float().sum(0, keepdim=True) res.append(correct_k.mul_(100.0 / batch_size)) return res def validate(val_loader, model, criterion): batch_time = AverageMeter() losses = AverageMeter() top1 = AverageMeter() top5 = AverageMeter() # switch to evaluate mode model.eval() with torch.no_grad(): end = time.time() for i, (input, target) in enumerate(val_loader): if args.gpu is not None: input = input.cuda(args.gpu, non_blocking=True) target = target.cuda(args.gpu, non_blocking=True) # compute output output = model(input) loss = criterion(output, target) # measure accuracy and record loss prec1, prec5 = accuracy(output, target, topk=(1, 5)) losses.update(loss.item(), input.size(0)) top1.update(prec1[0], input.size(0)) top5.update(prec5[0], input.size(0)) # measure elapsed time batch_time.update(time.time() - end) end = time.time() if i % args.print_freq == 0: print('Test: [{0}/{1}]\t' 'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format( i, len(val_loader), batch_time=batch_time, loss=losses, top1=top1, top5=top5)) print(' * Prec@1 {top1.avg:.3f} Prec@5 {top5.avg:.3f}' .format(top1=top1, top5=top5)) return top1.avg, top5.avg
null
22,958
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import models def save_checkpoint(state, is_best, filename='checkpoint.pth.tar'): directory = "runs/%s/"%(args.arch + '_' + args.action) filename = directory + filename torch.save(state, filename) if is_best: shutil.copyfile(filename, directory + 'model_best.pth.tar')
null
22,959
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import models The provided code snippet includes necessary dependencies for implementing the `adjust_learning_rate` function. Write a Python function `def adjust_learning_rate(optimizer, epoch)` to solve the following problem: Sets the learning rate to the initial LR decayed by 10 every 30 epochs Here is the function: def adjust_learning_rate(optimizer, epoch): """Sets the learning rate to the initial LR decayed by 10 every 30 epochs""" lr = args.lr * (0.1 ** (epoch // 30)) for param_group in optimizer.param_groups: param_group['lr'] = lr
Sets the learning rate to the initial LR decayed by 10 every 30 epochs
22,960
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import models def data_save(root, file): if not os.path.exists(root): os.mknod(root) file_temp = open(root, 'r') lines = file_temp.readlines() if not lines: epoch = -1 else: epoch = lines[-1][:lines[-1].index(' ')] epoch = int(epoch) file_temp.close() file_temp = open(root, 'a') for line in file: if line > epoch: file_temp.write(str(line) + " " + str(file[line]) + '\n') file_temp.close()
null
22,961
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import torchvision.models as models import models from thop import profile class AverageMeter(object): """Computes and stores the average and current value""" def __init__(self): self.reset() def reset(self): self.val = 0 self.avg = 0 self.sum = 0 self.count = 0 def update(self, val, n=1): self.val = val self.sum += val * n self.count += n self.avg = self.sum / self.count def accuracy(output, target, topk=(1,)): """Computes the precision@k for the specified values of k""" with torch.no_grad(): maxk = max(topk) batch_size = target.size(0) _, pred = output.topk(maxk, 1, True, True) pred = pred.t() correct = pred.eq(target.view(1, -1).expand_as(pred)) res = [] for k in topk: correct_k = correct[:k].view(-1).float().sum(0, keepdim=True) res.append(correct_k.mul_(100.0 / batch_size)) return res def train(train_loader, model, criterion, optimizer, epoch): batch_time = AverageMeter() data_time = AverageMeter() losses = AverageMeter() top1 = AverageMeter() top5 = AverageMeter() losses_batch = {} # switch to train mode model.train() end = time.time() for i, (input, target) in enumerate(train_loader): # measure data loading time data_time.update(time.time() - end) if args.gpu is not None: input = input.cuda(args.gpu, non_blocking=True) target = target.cuda(args.gpu, non_blocking=True) # compute output output = model(input) loss = criterion(output, target) # measure accuracy and record loss prec1, prec5 = accuracy(output, target, topk=(1, 5)) losses.update(loss.item(), input.size(0)) top1.update(prec1[0], input.size(0)) top5.update(prec5[0], input.size(0)) # compute gradient and do SGD step optimizer.zero_grad() loss.backward() optimizer.step() # measure elapsed time batch_time.update(time.time() - end) end = time.time() if i % args.print_freq == 0: print('Epoch: [{0}][{1}/{2}]\t' 'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t' 'Data {data_time.val:.3f} ({data_time.avg:.3f})\t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format( epoch, i, len(train_loader), batch_time=batch_time, data_time=data_time, loss=losses, top1=top1, top5=top5)) losses_batch[epoch] = losses.avg loss_batch = open("loss_batch.txt", 'a') for line in losses_batch: loss_batch.write(str(epoch) + " " + str(line) + " " + str(losses_batch[line]) + '\n') loss_batch.close() return losses.avg, top1.avg, top5.avg
null
22,962
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import torchvision.models as models import models from thop import profile class AverageMeter(object): def __init__(self): def reset(self): def update(self, val, n=1): def accuracy(output, target, topk=(1,)): def validate(val_loader, model, criterion): batch_time = AverageMeter() losses = AverageMeter() top1 = AverageMeter() top5 = AverageMeter() # switch to evaluate mode model.eval() with torch.no_grad(): end = time.time() for i, (input, target) in enumerate(val_loader): if args.gpu is not None: input = input.cuda(args.gpu, non_blocking=True) target = target.cuda(args.gpu, non_blocking=True) # compute output output = model(input) loss = criterion(output, target) # measure accuracy and record loss prec1, prec5 = accuracy(output, target, topk=(1, 5)) losses.update(loss.item(), input.size(0)) top1.update(prec1[0], input.size(0)) top5.update(prec5[0], input.size(0)) # measure elapsed time batch_time.update(time.time() - end) end = time.time() if i % args.print_freq == 0: print('Test: [{0}/{1}]\t' 'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format( i, len(val_loader), batch_time=batch_time, loss=losses, top1=top1, top5=top5)) print(' * Prec@1 {top1.avg:.3f} Prec@5 {top5.avg:.3f}' .format(top1=top1, top5=top5)) return top1.avg, top5.avg
null
22,963
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import torchvision.models as models import models from thop import profile def save_checkpoint(state, is_best, filename='checkpoint.pth.tar'): directory = "runs/%s/"%(args.arch + '_' + args.action) filename = directory + filename torch.save(state, filename) if is_best: shutil.copyfile(filename, directory + 'model_best.pth.tar')
null
22,964
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import torchvision.models as models import models from thop import profile The provided code snippet includes necessary dependencies for implementing the `adjust_learning_rate` function. Write a Python function `def adjust_learning_rate(optimizer, epoch)` to solve the following problem: Sets the learning rate to the initial LR decayed by 10 every 30 epochs Here is the function: def adjust_learning_rate(optimizer, epoch): """Sets the learning rate to the initial LR decayed by 10 every 30 epochs""" # lr = args.lr * (0.1 ** (epoch // 30)) lr = args.lr * (0.98 ** epoch) print('lr = ', lr) for param_group in optimizer.param_groups: param_group['lr'] = lr
Sets the learning rate to the initial LR decayed by 10 every 30 epochs
22,965
import argparse import os import random import shutil import time import warnings import torch import torch.nn as nn import torch.nn.parallel import torch.backends.cudnn as cudnn import torch.distributed as dist import torch.optim import torch.utils.data import torch.utils.data.distributed import torchvision.transforms as transforms import torchvision.datasets as datasets import torchvision.models as models import models from thop import profile def data_save(root, file): if not os.path.exists(root): os.mknod(root) file_temp = open(root, 'r') lines = file_temp.readlines() if not lines: epoch = -1 else: epoch = lines[-1][:lines[-1].index(' ')] epoch = int(epoch) file_temp.close() file_temp = open(root, 'a') for line in file: if line > epoch: file_temp.write(str(line) + " " + str(file[line]) + '\n') file_temp.close()
null
22,966
from torch import nn from .eca_module import eca_layer class ECA_MobileNetV2(nn.Module): def __init__(self, num_classes=1000, width_mult=1.0): super(ECA_MobileNetV2, self).__init__() block = InvertedResidual input_channel = 32 last_channel = 1280 inverted_residual_setting = [ # t, c, n, s [1, 16, 1, 1], [6, 24, 2, 2], [6, 32, 3, 2], [6, 64, 4, 2], [6, 96, 3, 1], [6, 160, 3, 2], [6, 320, 1, 1], ] # building first layer input_channel = int(input_channel * width_mult) self.last_channel = int(last_channel * max(1.0, width_mult)) features = [ConvBNReLU(3, input_channel, stride=2)] # building inverted residual blocks for t, c, n, s in inverted_residual_setting: output_channel = int(c * width_mult) for i in range(n): if c < 96: ksize = 1 else: ksize = 3 stride = s if i == 0 else 1 features.append(block(input_channel, output_channel, stride, expand_ratio=t, k_size=ksize)) input_channel = output_channel # building last several layers features.append(ConvBNReLU(input_channel, self.last_channel, kernel_size=1)) # make it nn.Sequential self.features = nn.Sequential(*features) # building classifier self.classifier = nn.Sequential( nn.Dropout(0.25), nn.Linear(self.last_channel, num_classes), ) # weight initialization for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode='fan_out') if m.bias is not None: nn.init.zeros_(m.bias) elif isinstance(m, nn.BatchNorm2d): nn.init.ones_(m.weight) nn.init.zeros_(m.bias) elif isinstance(m, nn.Linear): nn.init.normal_(m.weight, 0, 0.01) if m.bias is not None: nn.init.zeros_(m.bias) def forward(self, x): x = self.features(x) x = x.mean(-1).mean(-1) x = self.classifier(x) return x The provided code snippet includes necessary dependencies for implementing the `eca_mobilenet_v2` function. Write a Python function `def eca_mobilenet_v2(pretrained=False, progress=True, **kwargs)` to solve the following problem: Constructs a ECA_MobileNetV2 architecture from Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr Here is the function: def eca_mobilenet_v2(pretrained=False, progress=True, **kwargs): """ Constructs a ECA_MobileNetV2 architecture from Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr """ model = ECA_MobileNetV2(**kwargs) # if pretrained: # state_dict = load_state_dict_from_url(model_urls['mobilenet_v2'], # progress=progress) # model.load_state_dict(state_dict) return model
Constructs a ECA_MobileNetV2 architecture from Args: pretrained (bool): If True, returns a model pre-trained on ImageNet progress (bool): If True, displays a progress bar of the download to stderr
22,967
import torch.nn as nn import math from .eca_module import eca_layer The provided code snippet includes necessary dependencies for implementing the `conv3x3` function. Write a Python function `def conv3x3(in_planes, out_planes, stride=1)` to solve the following problem: 3x3 convolution with padding Here is the function: def conv3x3(in_planes, out_planes, stride=1): """3x3 convolution with padding""" return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)
3x3 convolution with padding
22,968
import torch.nn as nn import math from .eca_module import eca_layer class ECABasicBlock(nn.Module): expansion = 1 def __init__(self, inplanes, planes, stride=1, downsample=None, k_size=3): super(ECABasicBlock, self).__init__() self.conv1 = conv3x3(inplanes, planes, stride) self.bn1 = nn.BatchNorm2d(planes) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3(planes, planes, 1) self.bn2 = nn.BatchNorm2d(planes) self.eca = eca_layer(planes, k_size) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.eca(out) if self.downsample is not None: residual = self.downsample(x) out += residual out = self.relu(out) return out class ResNet(nn.Module): def __init__(self, block, layers, num_classes=1000, k_size=[3, 3, 3, 3]): self.inplanes = 64 super(ResNet, self).__init__() self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(64) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, layers[0], int(k_size[0])) self.layer2 = self._make_layer(block, 128, layers[1], int(k_size[1]), stride=2) self.layer3 = self._make_layer(block, 256, layers[2], int(k_size[2]), stride=2) self.layer4 = self._make_layer(block, 512, layers[3], int(k_size[3]), stride=2) self.avgpool = nn.AvgPool2d(7, stride=1) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels m.weight.data.normal_(0, math.sqrt(2. / n)) elif isinstance(m, nn.BatchNorm2d): m.weight.data.fill_(1) m.bias.data.zero_() def _make_layer(self, block, planes, blocks, k_size, stride=1): downsample = None if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, k_size)) self.inplanes = planes * block.expansion for i in range(1, blocks): layers.append(block(self.inplanes, planes, k_size=k_size)) return nn.Sequential(*layers) def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) x = self.layer1(x) x = self.layer2(x) x = self.layer3(x) x = self.layer4(x) x = self.avgpool(x) x = x.view(x.size(0), -1) x = self.fc(x) return x The provided code snippet includes necessary dependencies for implementing the `eca_resnet18` function. Write a Python function `def eca_resnet18(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False)` to solve the following problem: Constructs a ResNet-18 model. Args: k_size: Adaptive selection of kernel size pretrained (bool): If True, returns a model pre-trained on ImageNet num_classes:The classes of classification Here is the function: def eca_resnet18(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False): """Constructs a ResNet-18 model. Args: k_size: Adaptive selection of kernel size pretrained (bool): If True, returns a model pre-trained on ImageNet num_classes:The classes of classification """ model = ResNet(ECABasicBlock, [2, 2, 2, 2], num_classes=num_classes, k_size=k_size) model.avgpool = nn.AdaptiveAvgPool2d(1) return model
Constructs a ResNet-18 model. Args: k_size: Adaptive selection of kernel size pretrained (bool): If True, returns a model pre-trained on ImageNet num_classes:The classes of classification
22,969
import torch.nn as nn import math from .eca_module import eca_layer class ECABasicBlock(nn.Module): expansion = 1 def __init__(self, inplanes, planes, stride=1, downsample=None, k_size=3): super(ECABasicBlock, self).__init__() self.conv1 = conv3x3(inplanes, planes, stride) self.bn1 = nn.BatchNorm2d(planes) self.relu = nn.ReLU(inplace=True) self.conv2 = conv3x3(planes, planes, 1) self.bn2 = nn.BatchNorm2d(planes) self.eca = eca_layer(planes, k_size) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.eca(out) if self.downsample is not None: residual = self.downsample(x) out += residual out = self.relu(out) return out class ResNet(nn.Module): def __init__(self, block, layers, num_classes=1000, k_size=[3, 3, 3, 3]): self.inplanes = 64 super(ResNet, self).__init__() self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(64) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, layers[0], int(k_size[0])) self.layer2 = self._make_layer(block, 128, layers[1], int(k_size[1]), stride=2) self.layer3 = self._make_layer(block, 256, layers[2], int(k_size[2]), stride=2) self.layer4 = self._make_layer(block, 512, layers[3], int(k_size[3]), stride=2) self.avgpool = nn.AvgPool2d(7, stride=1) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels m.weight.data.normal_(0, math.sqrt(2. / n)) elif isinstance(m, nn.BatchNorm2d): m.weight.data.fill_(1) m.bias.data.zero_() def _make_layer(self, block, planes, blocks, k_size, stride=1): downsample = None if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, k_size)) self.inplanes = planes * block.expansion for i in range(1, blocks): layers.append(block(self.inplanes, planes, k_size=k_size)) return nn.Sequential(*layers) def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) x = self.layer1(x) x = self.layer2(x) x = self.layer3(x) x = self.layer4(x) x = self.avgpool(x) x = x.view(x.size(0), -1) x = self.fc(x) return x The provided code snippet includes necessary dependencies for implementing the `eca_resnet34` function. Write a Python function `def eca_resnet34(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False)` to solve the following problem: Constructs a ResNet-34 model. Args: k_size: Adaptive selection of kernel size pretrained (bool): If True, returns a model pre-trained on ImageNet num_classes:The classes of classification Here is the function: def eca_resnet34(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False): """Constructs a ResNet-34 model. Args: k_size: Adaptive selection of kernel size pretrained (bool): If True, returns a model pre-trained on ImageNet num_classes:The classes of classification """ model = ResNet(ECABasicBlock, [3, 4, 6, 3], num_classes=num_classes, k_size=k_size) model.avgpool = nn.AdaptiveAvgPool2d(1) return model
Constructs a ResNet-34 model. Args: k_size: Adaptive selection of kernel size pretrained (bool): If True, returns a model pre-trained on ImageNet num_classes:The classes of classification
22,970
import torch.nn as nn import math from .eca_module import eca_layer class ECABottleneck(nn.Module): expansion = 4 def __init__(self, inplanes, planes, stride=1, downsample=None, k_size=3): super(ECABottleneck, self).__init__() self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) self.bn1 = nn.BatchNorm2d(planes) self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(planes) self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False) self.bn3 = nn.BatchNorm2d(planes * 4) self.relu = nn.ReLU(inplace=True) self.eca = eca_layer(planes * 4, k_size) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.relu(out) out = self.conv3(out) out = self.bn3(out) out = self.eca(out) if self.downsample is not None: residual = self.downsample(x) out += residual out = self.relu(out) return out class ResNet(nn.Module): def __init__(self, block, layers, num_classes=1000, k_size=[3, 3, 3, 3]): self.inplanes = 64 super(ResNet, self).__init__() self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(64) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, layers[0], int(k_size[0])) self.layer2 = self._make_layer(block, 128, layers[1], int(k_size[1]), stride=2) self.layer3 = self._make_layer(block, 256, layers[2], int(k_size[2]), stride=2) self.layer4 = self._make_layer(block, 512, layers[3], int(k_size[3]), stride=2) self.avgpool = nn.AvgPool2d(7, stride=1) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels m.weight.data.normal_(0, math.sqrt(2. / n)) elif isinstance(m, nn.BatchNorm2d): m.weight.data.fill_(1) m.bias.data.zero_() def _make_layer(self, block, planes, blocks, k_size, stride=1): downsample = None if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, k_size)) self.inplanes = planes * block.expansion for i in range(1, blocks): layers.append(block(self.inplanes, planes, k_size=k_size)) return nn.Sequential(*layers) def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) x = self.layer1(x) x = self.layer2(x) x = self.layer3(x) x = self.layer4(x) x = self.avgpool(x) x = x.view(x.size(0), -1) x = self.fc(x) return x The provided code snippet includes necessary dependencies for implementing the `eca_resnet50` function. Write a Python function `def eca_resnet50(k_size=[3, 3, 3, 3], num_classes=1000, pretrained=False)` to solve the following problem: Constructs a ResNet-50 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet Here is the function: def eca_resnet50(k_size=[3, 3, 3, 3], num_classes=1000, pretrained=False): """Constructs a ResNet-50 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet """ print("Constructing eca_resnet50......") model = ResNet(ECABottleneck, [3, 4, 6, 3], num_classes=num_classes, k_size=k_size) model.avgpool = nn.AdaptiveAvgPool2d(1) return model
Constructs a ResNet-50 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet
22,971
import torch.nn as nn import math from .eca_module import eca_layer class ECABottleneck(nn.Module): expansion = 4 def __init__(self, inplanes, planes, stride=1, downsample=None, k_size=3): super(ECABottleneck, self).__init__() self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) self.bn1 = nn.BatchNorm2d(planes) self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(planes) self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False) self.bn3 = nn.BatchNorm2d(planes * 4) self.relu = nn.ReLU(inplace=True) self.eca = eca_layer(planes * 4, k_size) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.relu(out) out = self.conv3(out) out = self.bn3(out) out = self.eca(out) if self.downsample is not None: residual = self.downsample(x) out += residual out = self.relu(out) return out class ResNet(nn.Module): def __init__(self, block, layers, num_classes=1000, k_size=[3, 3, 3, 3]): self.inplanes = 64 super(ResNet, self).__init__() self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(64) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, layers[0], int(k_size[0])) self.layer2 = self._make_layer(block, 128, layers[1], int(k_size[1]), stride=2) self.layer3 = self._make_layer(block, 256, layers[2], int(k_size[2]), stride=2) self.layer4 = self._make_layer(block, 512, layers[3], int(k_size[3]), stride=2) self.avgpool = nn.AvgPool2d(7, stride=1) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels m.weight.data.normal_(0, math.sqrt(2. / n)) elif isinstance(m, nn.BatchNorm2d): m.weight.data.fill_(1) m.bias.data.zero_() def _make_layer(self, block, planes, blocks, k_size, stride=1): downsample = None if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, k_size)) self.inplanes = planes * block.expansion for i in range(1, blocks): layers.append(block(self.inplanes, planes, k_size=k_size)) return nn.Sequential(*layers) def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) x = self.layer1(x) x = self.layer2(x) x = self.layer3(x) x = self.layer4(x) x = self.avgpool(x) x = x.view(x.size(0), -1) x = self.fc(x) return x The provided code snippet includes necessary dependencies for implementing the `eca_resnet101` function. Write a Python function `def eca_resnet101(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False)` to solve the following problem: Constructs a ResNet-101 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet Here is the function: def eca_resnet101(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False): """Constructs a ResNet-101 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet """ model = ResNet(ECABottleneck, [3, 4, 23, 3], num_classes=num_classes, k_size=k_size) model.avgpool = nn.AdaptiveAvgPool2d(1) return model
Constructs a ResNet-101 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet
22,972
import torch.nn as nn import math from .eca_module import eca_layer class ECABottleneck(nn.Module): expansion = 4 def __init__(self, inplanes, planes, stride=1, downsample=None, k_size=3): super(ECABottleneck, self).__init__() self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) self.bn1 = nn.BatchNorm2d(planes) self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(planes) self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False) self.bn3 = nn.BatchNorm2d(planes * 4) self.relu = nn.ReLU(inplace=True) self.eca = eca_layer(planes * 4, k_size) self.downsample = downsample self.stride = stride def forward(self, x): residual = x out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) out = self.relu(out) out = self.conv3(out) out = self.bn3(out) out = self.eca(out) if self.downsample is not None: residual = self.downsample(x) out += residual out = self.relu(out) return out class ResNet(nn.Module): def __init__(self, block, layers, num_classes=1000, k_size=[3, 3, 3, 3]): self.inplanes = 64 super(ResNet, self).__init__() self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(64) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, layers[0], int(k_size[0])) self.layer2 = self._make_layer(block, 128, layers[1], int(k_size[1]), stride=2) self.layer3 = self._make_layer(block, 256, layers[2], int(k_size[2]), stride=2) self.layer4 = self._make_layer(block, 512, layers[3], int(k_size[3]), stride=2) self.avgpool = nn.AvgPool2d(7, stride=1) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels m.weight.data.normal_(0, math.sqrt(2. / n)) elif isinstance(m, nn.BatchNorm2d): m.weight.data.fill_(1) m.bias.data.zero_() def _make_layer(self, block, planes, blocks, k_size, stride=1): downsample = None if stride != 1 or self.inplanes != planes * block.expansion: downsample = nn.Sequential( nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(planes * block.expansion), ) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, k_size)) self.inplanes = planes * block.expansion for i in range(1, blocks): layers.append(block(self.inplanes, planes, k_size=k_size)) return nn.Sequential(*layers) def forward(self, x): x = self.conv1(x) x = self.bn1(x) x = self.relu(x) x = self.maxpool(x) x = self.layer1(x) x = self.layer2(x) x = self.layer3(x) x = self.layer4(x) x = self.avgpool(x) x = x.view(x.size(0), -1) x = self.fc(x) return x The provided code snippet includes necessary dependencies for implementing the `eca_resnet152` function. Write a Python function `def eca_resnet152(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False)` to solve the following problem: Constructs a ResNet-152 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet Here is the function: def eca_resnet152(k_size=[3, 3, 3, 3], num_classes=1_000, pretrained=False): """Constructs a ResNet-152 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet """ model = ResNet(ECABottleneck, [3, 8, 36, 3], num_classes=num_classes, k_size=k_size) model.avgpool = nn.AdaptiveAvgPool2d(1) return model
Constructs a ResNet-152 model. Args: k_size: Adaptive selection of kernel size num_classes:The classes of classification pretrained (bool): If True, returns a model pre-trained on ImageNet
22,973
import os import sys from setuptools import setup, find_packages, dist import glob import logging import subprocess import torch from torch.utils.cpp_extension import BuildExtension, CppExtension, CUDAExtension, CUDA_HOME def get_cuda_bare_metal_version(cuda_dir): raw_output = subprocess.check_output([cuda_dir + "/bin/nvcc", "-V"], universal_newlines=True) output = raw_output.split() release_idx = output.index("release") + 1 release = output[release_idx].split(".") bare_metal_major = release[0] bare_metal_minor = release[1][0] return raw_output, bare_metal_major, bare_metal_minor
null
22,974
import os import sys from setuptools import setup, find_packages, dist import glob import logging import subprocess import torch from torch.utils.cpp_extension import BuildExtension, CppExtension, CUDAExtension, CUDA_HOME if not torch.cuda.is_available(): if os.getenv('FORCE_CUDA', '0') == '1': # From: https://github.com/NVIDIA/apex/blob/c4e85f7bf144cb0e368da96d339a6cbd9882cea5/setup.py # Extension builds after https://github.com/pytorch/pytorch/pull/23408 attempt to query torch.cuda.get_device_capability(), # which will fail if you are compiling in an environment without visible GPUs (e.g. during an nvidia-docker build command). logging.warning( "Torch did not find available GPUs on this system.\n" "If your intention is to cross-compile, this is not an error.\n" "By default, Apex will cross-compile for Pascal (compute capabilities 6.0, 6.1, 6.2),\n" "Volta (compute capability 7.0), Turing (compute capability 7.5),\n" "and, if the CUDA version is >= 11.0, Ampere (compute capability 8.0).\n" "If you wish to cross-compile for a single specific architecture,\n" 'export TORCH_CUDA_ARCH_LIST="compute capability" before running setup.py.\n' ) if os.getenv("TORCH_CUDA_ARCH_LIST", None) is None: _, bare_metal_major, bare_metal_minor = get_cuda_bare_metal_version(CUDA_HOME) if int(bare_metal_major) == 11: if int(bare_metal_minor) == 0: os.environ["TORCH_CUDA_ARCH_LIST"] = "6.0;6.1;6.2;7.0;7.5;8.0" else: os.environ["TORCH_CUDA_ARCH_LIST"] = "6.0;6.1;6.2;7.0;7.5;8.0;8.6" else: os.environ["TORCH_CUDA_ARCH_LIST"] = "6.0;6.1;6.2;7.0;7.5" else: logging.warning( "Torch did not find available GPUs on this system.\n" "This script will install only with CPU support and will have very limited features.\n" 'If your wish to cross-compile for GPU `export FORCE_CUDA=1` before running setup.py\n' "By default, Apex will cross-compile for Pascal (compute capabilities 6.0, 6.1, 6.2),\n" "Volta (compute capability 7.0), Turing (compute capability 7.5),\n" "and, if the CUDA version is >= 11.0, Ampere (compute capability 8.0).\n" "If you wish to cross-compile for a single specific architecture,\n" 'export TORCH_CUDA_ARCH_LIST="compute capability" before running setup.py.\n' ) def get_extensions(): extra_compile_args = {'cxx': ['-O3', '-fdiagnostics-color=always']} define_macros = [] include_dirs = [] extensions = [] sources = glob.glob('wisp/csrc/**/*.cpp', recursive=True) if len(sources) == 0: print("No source files found for extension, skipping extension compilation") return None if torch.cuda.is_available() or os.getenv('FORCE_CUDA', '0') == '1': define_macros += [("WITH_CUDA", None), ("THRUST_IGNORE_CUB_VERSION_CHECK", None)] sources += glob.glob('wisp/csrc/**/*.cu', recursive=True) extension = CUDAExtension extra_compile_args.update({'nvcc': ['-O3']}) #include_dirs = get_include_dirs() else: assert(False, "CUDA is not available. Set FORCE_CUDA=1 for Docker builds") extensions.append( extension( name='wisp._C', sources=sources, define_macros=define_macros, extra_compile_args=extra_compile_args, #include_dirs=include_dirs ) ) for ext in extensions: ext.libraries = ['cudart_static' if x == 'cudart' else x for x in ext.libraries] return extensions
null
22,975
import os def setup(app): # -- To demonstrate ReadTheDocs switcher ------------------------------------- # This links a few JS and CSS files that mimic the environment that RTD uses # so that we can test RTD-like behavior. We don't need to run it on RTD and we # don't want it loaded in GitHub Actions if not os.environ.get("READTHEDOCS") and not os.environ.get("GITHUB_ACTIONS"): app.add_css_file( "https://assets.readthedocs.org/static/css/readthedocs-doc-embed.css" ) app.add_css_file("https://assets.readthedocs.org/static/css/badge_only.css") # Create the dummy data file so we can link it # ref: https://github.com/readthedocs/readthedocs.org/blob/bc3e147770e5740314a8e8c33fec5d111c850498/readthedocs/core/static-src/core/js/doc-embed/footer.js # noqa: E501 app.add_js_file("rtd-data.js") app.add_js_file( "https://assets.readthedocs.org/static/javascript/readthedocs-doc-embed.js", priority=501, )
null
22,976
import numpy as np import torch import pathlib import argparse from kaolin.io.obj import import_mesh from kaolin.ops.mesh import sample_points from kaolin.render.mesh.utils import texture_mapping from kaolin.ops.conversions.pointcloud import unbatched_pointcloud_to_spc def convert_texture_to_torch_sample_format(texture, sampled_uvs): """ Convert to (1, C, Tex-H, Tex-W) format """ return texture.unsqueeze(0).type(sampled_uvs.dtype).permute(0, 3, 1, 2) The provided code snippet includes necessary dependencies for implementing the `convert_mesh_to_spc` function. Write a Python function `def convert_mesh_to_spc(mesh_path, level, output_path, num_samples)` to solve the following problem: Loads obj and converts it to a SPC. Output will reside in output_path. Here is the function: def convert_mesh_to_spc(mesh_path, level, output_path, num_samples): """ Loads obj and converts it to a SPC. Output will reside in output_path.""" mesh = import_mesh(mesh_path, with_materials=True) print(f'Loaded mesh with {len(mesh.vertices)} vertices, {len(mesh.faces)} faces and {len(mesh.materials)} materials.') # Load the uv coordinates per face-vertex like "features" per face-vertex, # which sample_points will interpolate for new sample points. # mesh.uvs is a tensor of uv coordinates of shape (#num_uvs, 2), which we consider as "features" here # mesh.face_uvs_idx is a tensor of shape (#faces, 3), indexing which feature to use per-face-per-vertex # Therefore, face_features will be of shape (#faces, 3, 2) face_features = mesh.uvs[mesh.face_uvs_idx] # Kaolin assumes an exact batch format, we make sure to convert from: # (V, 3) to (1, V, 3) # (F, 3, 2) to (1, F, 3, 2) # where 1 is the batch size batched_vertices = mesh.vertices.unsqueeze(0) batched_face_features = face_features.unsqueeze(0) # sample_points is faster on cuda device batched_vertices = batched_vertices.cuda() faces = mesh.faces.cuda() batched_face_features = batched_face_features.cuda() sampled_verts, _, sampled_uvs = sample_points(batched_vertices, faces, num_samples=num_samples, face_features=batched_face_features) print(f'Sampled {sampled_verts.shape[1]} points over the mesh surface.') # Convert texture to sample-compatible format diffuse_color = mesh.materials[0]['map_Kd'] # Assumes a shape with a single material texture_maps = convert_texture_to_torch_sample_format(diffuse_color, sampled_uvs) # (1, C, Th, Tw) texture_maps = texture_maps.cuda() # Sample colors according to uv-coordinates sampled_uvs = texture_mapping(texture_coordinates=sampled_uvs, texture_maps=texture_maps, mode='nearest') # Unbatch vertices = sampled_verts.squeeze(0) vertex_colors = sampled_uvs.squeeze(0) # Normalize to [0,1], and convert to RGBA if needed vertex_colors /= 255 if vertex_colors.shape[-1] == 3: vertex_colors = torch.cat([vertex_colors, torch.ones_like(vertex_colors[:, :1])], dim=1) spc = unbatched_pointcloud_to_spc(vertices, level, features=vertex_colors) print(f'SPC generated with {spc.point_hierarchies.shape[0]} cells.') octrees_entry = spc.octrees colors_entry = (255 * spc.features.reshape(-1)) npz_record = dict( octree=octrees_entry.cpu().numpy().astype(np.uint8), colors=colors_entry.cpu().numpy().astype(np.uint8) ) # Default output path: take filename and save in current working directory if output_path is None: output_path = pathlib.Path(mesh_path).stem output_path = f'{output_path}.npz' np.savez(output_path, **npz_record) return output_path
Loads obj and converts it to a SPC. Output will reside in output_path.
22,977
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `identity` function. Write a Python function `def identity(c: torch.Tensor) -> torch.Tensor` to solve the following problem: A naive normalization function which assumes the value is already normalized and returned as is. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. Returns: (torch.Tensor): Input channel c is returned without a change. Here is the function: def identity(c: torch.Tensor) -> torch.Tensor: """ A naive normalization function which assumes the value is already normalized and returned as is. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. Returns: (torch.Tensor): Input channel c is returned without a change. """ return c
A naive normalization function which assumes the value is already normalized and returned as is. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. Returns: (torch.Tensor): Input channel c is returned without a change.
22,978
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F def normalize(c: torch.Tensor, min_val: Any = None, max_val: Any = None) -> torch.Tensor: """ A linear normalization function which maps the channel c to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1]. """ min_val = torch.min(c) if min_val is None else min_val max_val = torch.max(c) if max_val is None else max_val return (c - min_val) / (max_val - min_val) The provided code snippet includes necessary dependencies for implementing the `normalize_linear_scale` function. Write a Python function `def normalize_linear_scale(c: torch.Tensor, min_val: Any = None, max_val: Any = None, linear_scale: float = 1.0) -> torch.Tensor` to solve the following problem: A normalization function which linear scales the channel before normalizing it to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. If explicitly specified, the bounds are scaled as well. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C linear_scale (float): Channel scale value Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1]. Here is the function: def normalize_linear_scale(c: torch.Tensor, min_val: Any = None, max_val: Any = None, linear_scale: float = 1.0) -> torch.Tensor: """ A normalization function which linear scales the channel before normalizing it to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. If explicitly specified, the bounds are scaled as well. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C linear_scale (float): Channel scale value Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1]. """ c *= linear_scale min_val = linear_scale * min_val if min_val is not None else min_val max_val = linear_scale * max_val if max_val is not None else max_val return normalize(c=c, min_val=min_val, max_val=max_val)
A normalization function which linear scales the channel before normalizing it to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. If explicitly specified, the bounds are scaled as well. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C linear_scale (float): Channel scale value Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1].
22,979
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F def normalize(c: torch.Tensor, min_val: Any = None, max_val: Any = None) -> torch.Tensor: """ A linear normalization function which maps the channel c to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1]. """ min_val = torch.min(c) if min_val is None else min_val max_val = torch.max(c) if max_val is None else max_val return (c - min_val) / (max_val - min_val) The provided code snippet includes necessary dependencies for implementing the `normalize_log_scale` function. Write a Python function `def normalize_log_scale(c: torch.Tensor, min_val: Any = None, max_val: Any = None, linear_scale: float = 1.0, log_scale: float = 1.0) -> torch.Tensor` to solve the following problem: A normalization function which applies log and linear scales to the channel before normalizing it to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. If explicitly specified, the bounds are scaled as well. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C linear_scale (float): Channel linear scale value log_scale (float: Channel log scale value. Returns: (torch.Tensor): Input channel c is mapped to the range [0, 1] by the formula: ``c' = a * log(b * c)`` where a is the linear scale, b is the log scale and c is the channel value. Here is the function: def normalize_log_scale(c: torch.Tensor, min_val: Any = None, max_val: Any = None, linear_scale: float = 1.0, log_scale: float = 1.0) -> torch.Tensor: """ A normalization function which applies log and linear scales to the channel before normalizing it to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. If explicitly specified, the bounds are scaled as well. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C linear_scale (float): Channel linear scale value log_scale (float: Channel log scale value. Returns: (torch.Tensor): Input channel c is mapped to the range [0, 1] by the formula: ``c' = a * log(b * c)`` where a is the linear scale, b is the log scale and c is the channel value. """ c = linear_scale * torch.log(log_scale * c) min_val = linear_scale * torch.log(log_scale * min_val) if min_val is not None else min_val max_val = linear_scale * torch.log(log_scale * max_val) if max_val is not None else max_val return normalize(c=c, min_val=min_val, max_val=max_val)
A normalization function which applies log and linear scales to the channel before normalizing it to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. If explicitly specified, the bounds are scaled as well. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C linear_scale (float): Channel linear scale value log_scale (float: Channel log scale value. Returns: (torch.Tensor): Input channel c is mapped to the range [0, 1] by the formula: ``c' = a * log(b * c)`` where a is the linear scale, b is the log scale and c is the channel value.
22,980
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F def normalize(c: torch.Tensor, min_val: Any = None, max_val: Any = None) -> torch.Tensor: """ A linear normalization function which maps the channel c to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1]. """ min_val = torch.min(c) if min_val is None else min_val max_val = torch.max(c) if max_val is None else max_val return (c - min_val) / (max_val - min_val) The provided code snippet includes necessary dependencies for implementing the `normalize_vector` function. Write a Python function `def normalize_vector(c: torch.Tensor) -> torch.Tensor` to solve the following problem: A normalization function which applies a L2 normalization over a channel of vector data. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. Returns: (torch.Tensor): Input channel c is normalized by the L2 norm. Here is the function: def normalize_vector(c: torch.Tensor) -> torch.Tensor: """ A normalization function which applies a L2 normalization over a channel of vector data. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. Returns: (torch.Tensor): Input channel c is normalized by the L2 norm. """ return F.normalize(c, dim=-1) * 0.5 + 0.5
A normalization function which applies a L2 normalization over a channel of vector data. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. Returns: (torch.Tensor): Input channel c is normalized by the L2 norm.
22,981
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_linear` function. Write a Python function `def blend_linear(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: A direct linear interpolation between c1 and c2. Useful for blending channels which do not consider the alpha value (i.e. the alpha channel itself). Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused alpha2 (torch.Tensor): Unused Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_linear(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ A direct linear interpolation between c1 and c2. Useful for blending channels which do not consider the alpha value (i.e. the alpha channel itself). Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused alpha2 (torch.Tensor): Unused Returns: (torch.Tensor): Blended channel in the shape of c1 """ return c1 + c2 * (1.0 - c1)
A direct linear interpolation between c1 and c2. Useful for blending channels which do not consider the alpha value (i.e. the alpha channel itself). Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused alpha2 (torch.Tensor): Unused Returns: (torch.Tensor): Blended channel in the shape of c1
22,982
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_alpha_composite_over` function. Write a Python function `def blend_alpha_composite_over(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: An alpha compositing op where a front pixel is alpha blended with the background pixel (in a usual painter's algorithm manner). Useful for blending channels such as RGB. See: https://en.wikipedia.org/wiki/Alpha_compositing Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): alpha channel tensor, corresponding to second channel, in the shape of c1. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_alpha_composite_over(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ An alpha compositing op where a front pixel is alpha blended with the background pixel (in a usual painter's algorithm manner). Useful for blending channels such as RGB. See: https://en.wikipedia.org/wiki/Alpha_compositing Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): alpha channel tensor, corresponding to second channel, in the shape of c1. Returns: (torch.Tensor): Blended channel in the shape of c1 """ alpha_out = alpha1 + alpha2 * (1.0 - alpha1) c_out = torch.where(condition=alpha_out > 0, input=(c1 * alpha1 + c2 * alpha2 * (1.0 - alpha1)) / alpha_out, other=torch.zeros_like(c1)) return c_out
An alpha compositing op where a front pixel is alpha blended with the background pixel (in a usual painter's algorithm manner). Useful for blending channels such as RGB. See: https://en.wikipedia.org/wiki/Alpha_compositing Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): alpha channel tensor, corresponding to second channel, in the shape of c1. Returns: (torch.Tensor): Blended channel in the shape of c1
22,983
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_alpha_lerp` function. Write a Python function `def blend_alpha_lerp(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: A linear interpolation between c1 and c2, which uses the alpha channel as a weighting factor. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_alpha_lerp(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ A linear interpolation between c1 and c2, which uses the alpha channel as a weighting factor. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return c1 * alpha1 + c2 * (1.0 - alpha1)
A linear interpolation between c1 and c2, which uses the alpha channel as a weighting factor. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,984
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F def normalize(c: torch.Tensor, min_val: Any = None, max_val: Any = None) -> torch.Tensor: """ A linear normalization function which maps the channel c to the range of [0, 1]. If the min / max values bounds of the channel are not explicitly specified, they're determined by c's values. Args: c (torch.Tensor): A single channel tensor of an arbitrary shape. min_val (Any): Optional, the lower boundary for possible values in channel C max_val (Any): Optional, the upper boundary for possible values in channel C Returns: (torch.Tensor): Input channel c is linearly mapped to the range [0, 1]. """ min_val = torch.min(c) if min_val is None else min_val max_val = torch.max(c) if max_val is None else max_val return (c - min_val) / (max_val - min_val) The provided code snippet includes necessary dependencies for implementing the `blend_alpha_slerp` function. Write a Python function `def blend_alpha_slerp(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: A spherical linear interpolation, useful for interpolating rotations or blending directional vectors. c1 and c2 are normalized and interpolated over the unit hypersphere. alpha1 acts as the interpolation weight. See: https://en.wikipedia.org/wiki/Slerp Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_alpha_slerp(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ A spherical linear interpolation, useful for interpolating rotations or blending directional vectors. c1 and c2 are normalized and interpolated over the unit hypersphere. alpha1 acts as the interpolation weight. See: https://en.wikipedia.org/wiki/Slerp Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ # To unit directions t = alpha1 # alpha1 is used as the interpolation weight c1 = F.normalize(c1, dim=1) c2 = F.normalize(c2, dim=1) dot = (c1*c2).sum(1) # batched dot prod omega = torch.acos(dot) # angle between directions sin_omega = torch.sin(omega) # TODO (operel): Be careful of omega=0.0 case c2_weight = (torch.sin((1.0 - t) * omega) / sin_omega).unsqueeze(1) c1_weight = (torch.sin(t * omega) / sin_omega).unsqueeze(1) res = c2_weight * c2 + c1_weight * c1 return res
A spherical linear interpolation, useful for interpolating rotations or blending directional vectors. c1 and c2 are normalized and interpolated over the unit hypersphere. alpha1 acts as the interpolation weight. See: https://en.wikipedia.org/wiki/Slerp Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): alpha channel tensor, corresponding to first channel, in the shape of c1. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,985
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_normal` function. Write a Python function `def blend_normal(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: A standard blend mode which uses the front pixel value, without mixing. Useful when alpha blending is undesired, or the channel contains categorical info (i.e. semantic class ids). Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): Unused alpha1 (torch.Tensor): Unused alpha2 (torch.Tensor): Unused Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_normal(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ A standard blend mode which uses the front pixel value, without mixing. Useful when alpha blending is undesired, or the channel contains categorical info (i.e. semantic class ids). Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): Unused alpha1 (torch.Tensor): Unused alpha2 (torch.Tensor): Unused Returns: (torch.Tensor): Blended channel in the shape of c1 """ return c1
A standard blend mode which uses the front pixel value, without mixing. Useful when alpha blending is undesired, or the channel contains categorical info (i.e. semantic class ids). Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): Unused alpha1 (torch.Tensor): Unused alpha2 (torch.Tensor): Unused Returns: (torch.Tensor): Blended channel in the shape of c1
22,986
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_multiply` function. Write a Python function `def blend_multiply(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: Commutative blend mode which preserves dark colors. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_multiply(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ Commutative blend mode which preserves dark colors. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return c1 * c2
Commutative blend mode which preserves dark colors. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,987
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_screen` function. Write a Python function `def blend_screen(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: Commutative blend mode which preserves light colors. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_screen(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ Commutative blend mode which preserves light colors. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return 1.0 - (1.0 - c1)(1.0 - c2)
Commutative blend mode which preserves light colors. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,988
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_add` function. Write a Python function `def blend_add(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: An additive blend mode, for aggregation of channel information. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_add(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ An additive blend mode, for aggregation of channel information. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return c1 + c2
An additive blend mode, for aggregation of channel information. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,989
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_sub` function. Write a Python function `def blend_sub(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: An subtractive blend mode, for removing channel information. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_sub(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ An subtractive blend mode, for removing channel information. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return c1 - c2
An subtractive blend mode, for removing channel information. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,990
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_logical_and` function. Write a Python function `def blend_logical_and(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: For boolean channels, blends with a logical AND function. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_logical_and(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ For boolean channels, blends with a logical AND function. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return torch.logical_and(c1, c2).to(c1.dtype)
For boolean channels, blends with a logical AND function. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,991
from __future__ import annotations from typing import Callable, Any import torch import torch.nn.functional as F The provided code snippet includes necessary dependencies for implementing the `blend_logical_or` function. Write a Python function `def blend_logical_or(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor)` to solve the following problem: For boolean channels, blends with a logical OR function. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 Here is the function: def blend_logical_or(c1: torch.Tensor, c2: torch.Tensor, alpha1: torch.Tensor, alpha2: torch.Tensor): """ For boolean channels, blends with a logical OR function. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1 """ return torch.logical_or(c1, c2).to(c1.dtype)
For boolean channels, blends with a logical OR function. Args: c1 (torch.Tensor): first channel tensor of an arbitrary shape. c2 (torch.Tensor): second channel tensor, in the shape of c1. alpha1 (torch.Tensor): Unused. alpha2 (torch.Tensor): Unused. Returns: (torch.Tensor): Blended channel in the shape of c1
22,992
from __future__ import annotations from wisp.core.channel_fn import * from dataclasses import dataclass from typing import Any, Optional, Dict from functools import partial class Channel: """ Defines how a Renderbuffer channel should behave in terms of functionalities like blending, normalization, and boundaries. """ blend_fn: BlendFunction = None """" How to blend information from this channel between 2 RenderBuffers """ normalize_fn: NormalizeFunction = normalize """ How to normalize the channel information to a scale of [0, 1] """ min_val: Optional[Any] = None """ Minimal valid value supported by this channel type. None indicates the valid values range from -inf. """ max_val: Optional[Any] = None """ Maximal valid value supported by this channel type . None indicates the valid values range to inf. """ The provided code snippet includes necessary dependencies for implementing the `create_default_channel` function. Write a Python function `def create_default_channel() -> Channel` to solve the following problem: A general channel template, to be used if no information about a channel have been recorded Here is the function: def create_default_channel() -> Channel: """ A general channel template, to be used if no information about a channel have been recorded """ return Channel( blend_fn=blend_alpha_composite_over, normalize_fn=normalize, min_val=None, max_val=None )
A general channel template, to be used if no information about a channel have been recorded
22,993
from __future__ import annotations from wisp.core.channel_fn import * from dataclasses import dataclass from typing import Any, Optional, Dict from functools import partial class Channel: """ Defines how a Renderbuffer channel should behave in terms of functionalities like blending, normalization, and boundaries. """ blend_fn: BlendFunction = None """" How to blend information from this channel between 2 RenderBuffers """ normalize_fn: NormalizeFunction = normalize """ How to normalize the channel information to a scale of [0, 1] """ min_val: Optional[Any] = None """ Minimal valid value supported by this channel type. None indicates the valid values range from -inf. """ max_val: Optional[Any] = None """ Maximal valid value supported by this channel type . None indicates the valid values range to inf. """ The provided code snippet includes necessary dependencies for implementing the `channels_starter_kit` function. Write a Python function `def channels_starter_kit() -> Dict[str, Channel]` to solve the following problem: Creates a predefined kit of channels commonly useful in the context of Wisp. Users may augment or replace this kit with additional custom channels. Here is the function: def channels_starter_kit() -> Dict[str, Channel]: """ Creates a predefined kit of channels commonly useful in the context of Wisp. Users may augment or replace this kit with additional custom channels. """ return dict( rgb=Channel( blend_fn=blend_alpha_composite_over, normalize_fn=identity, min_val=0.0, max_val=1.0 ), alpha=Channel( blend_fn=blend_linear, normalize_fn=normalize, min_val=0.0, max_val=1.0 ), depth=Channel( blend_fn=blend_normal, normalize_fn=partial(normalize_linear_scale, linear_scale=1000.0), min_val=0.0 ), normal=Channel( blend_fn=blend_alpha_slerp, normalize_fn=normalize_vector ), hit=Channel( blend_fn=blend_logical_or, normalize_fn=identity ), err=Channel( blend_fn=blend_add, normalize_fn=normalize ), gt=Channel( blend_fn=blend_alpha_composite_over, normalize_fn=identity, min_val=0.0, max_val=1.0 ) )
Creates a predefined kit of channels commonly useful in the context of Wisp. Users may augment or replace this kit with additional custom channels.
22,994
from typing import List, Tuple def color_wheel(): """ Returns: (list) a list of all colors defined in the color module. Each entry is a tuple of 3 floats (RGB values). """ return [ white, black, dark_gray, light_purple, lime, red, green, blue, orange, light_cyan, light_pink, light_yellow, light_teal, gray, soft_blue, soft_red, lime_green, purple, gold ] The provided code snippet includes necessary dependencies for implementing the `colors_generator` function. Write a Python function `def colors_generator(skip_colors: List = None) -> Tuple[float, float, float]` to solve the following problem: Generates the next color in the color wheel on each invocation. This generator repeats the color wheel cyclically when exhausted. Args: skip_colors Here is the function: def colors_generator(skip_colors: List = None) -> Tuple[float, float, float]: """ Generates the next color in the color wheel on each invocation. This generator repeats the color wheel cyclically when exhausted. Args: skip_colors """ if skip_colors is None: skip_colors = [] while True: for color in color_wheel(): if color in skip_colors: continue yield color
Generates the next color in the color wheel on each invocation. This generator repeats the color wheel cyclically when exhausted. Args: skip_colors
22,995
from __future__ import annotations import time import numpy as np import torch import torch.nn.functional as F from typing import Tuple from wisp.core import RenderBuffer, Rays from wisp.ops.shaders import matcap_shader, pointlight_shadow_shader from wisp.ops.differential import finitediff_gradient from wisp.ops.geometric import normalized_grid, normalized_slice def _generate_rays(camera_origin, camera_view, camera_right, camera_up, height, width, mode='persp', fov=90.0, device='cuda'): """Generates rays from camera parameters. Args: camera_origin (torch.FloatTensor): [3] size tensor specifying the camera origin camera_view (torch.FloatTensor): [3] size tensor specifying the camera view direction camera_right (torch.FloatTensor): [3] size tensor specifying the camera right direction camera_up (torch.FloatTensor): [3] size tensor specifying the camera up direction height (int): height of the image width (int): width of the image mode (str): string that specifies the camera mode. fov (float): field of view of the camera. device (str): device the tensor will be allocated on. Returns: (torch.FloatTensor, torch.FloatTensor): - [height, width, 3] tensor of ray origins - [height, width, 3] tensor of ray directions """ coord = normalized_grid(height, width, device=device) ray_origin = camera_right * coord[..., 0, np.newaxis] * np.tan(np.radians(fov / 2)) + \ camera_up * coord[..., 1, np.newaxis] * np.tan(np.radians(fov / 2)) + \ camera_origin + camera_view ray_origin = ray_origin.reshape(-1, 3) ray_offset = camera_view.unsqueeze(0).repeat(ray_origin.shape[0], 1) if mode == 'ortho': # Orthographic camera ray_dir = F.normalize(ray_offset, dim=-1) elif mode == 'persp': # Perspective camera ray_dir = F.normalize(ray_origin - camera_origin, dim=-1) ray_origin = camera_origin.repeat(ray_dir.shape[0], 1) else: raise ValueError('Invalid camera mode!') return ray_origin, ray_dir The provided code snippet includes necessary dependencies for implementing the `_look_at` function. Write a Python function `def _look_at(f, t, height, width, mode='persp', fov=90.0, device='cuda')` to solve the following problem: Vectorized look-at function, returns an array of ray origins and directions This function is mostly just a wrapper on top of generate_rays, but will calculate for you the view, right, and up vectors based on the from and to. Args: f (list of floats): [3] size list or tensor specifying the camera origin t (list of floats): [3] size list or tensor specifying the camera look at point height (int): height of the image width (int): width of the image mode (str): string that specifies the camera mode. fov (float): field of view of the camera. device (str): device the tensor will be allocated on. Returns: (torch.FloatTensor, torch.FloatTensor): - [height, width, 3] tensor of ray origins - [height, width, 3] tensor of ray directions Here is the function: def _look_at(f, t, height, width, mode='persp', fov=90.0, device='cuda'): """Vectorized look-at function, returns an array of ray origins and directions This function is mostly just a wrapper on top of generate_rays, but will calculate for you the view, right, and up vectors based on the from and to. Args: f (list of floats): [3] size list or tensor specifying the camera origin t (list of floats): [3] size list or tensor specifying the camera look at point height (int): height of the image width (int): width of the image mode (str): string that specifies the camera mode. fov (float): field of view of the camera. device (str): device the tensor will be allocated on. Returns: (torch.FloatTensor, torch.FloatTensor): - [height, width, 3] tensor of ray origins - [height, width, 3] tensor of ray directions """ camera_origin = torch.FloatTensor(f).to(device) camera_view = F.normalize(torch.FloatTensor(t).to(device) - camera_origin, dim=0) camera_right = F.normalize(torch.cross(camera_view, torch.FloatTensor([0, 1, 0]).to(device)), dim=0) camera_up = F.normalize(torch.cross(camera_right, camera_view), dim=0) return _generate_rays(camera_origin, camera_view, camera_right, camera_up, height, width, mode=mode, fov=fov, device=device)
Vectorized look-at function, returns an array of ray origins and directions This function is mostly just a wrapper on top of generate_rays, but will calculate for you the view, right, and up vectors based on the from and to. Args: f (list of floats): [3] size list or tensor specifying the camera origin t (list of floats): [3] size list or tensor specifying the camera look at point height (int): height of the image width (int): width of the image mode (str): string that specifies the camera mode. fov (float): field of view of the camera. device (str): device the tensor will be allocated on. Returns: (torch.FloatTensor, torch.FloatTensor): - [height, width, 3] tensor of ray origins - [height, width, 3] tensor of ray directions
22,996
import torch from wisp.core import ObjectTransform from wisp.models import Pipeline, RasterizationPipeline from wisp.framework import WispState, BottomLevelRendererState def add_pipeline_to_scene_graph(state: WispState, name: str, pipeline: Pipeline, transform: ObjectTransform = None, **setup_args): """ Adds a new object pipeline to the scene graph. The pipeline contains all components required to trace this object. The object is represented in the SceneGraph by a newly created BottomLevelRenderer. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph pipeline (Pipeline): A pipeline consisting of a field (and possibly a tracer) of the added object. transform (ObjectTransform): The object transform, containing a 4x4 transformation matrix which transforms the object from local object space to world space. setup_args (Dict): Optional setup args which control how the BottomLevelRenderer will be created. """ if transform is None: transform = ObjectTransform(device=pipeline.nef.device) state.graph.neural_pipelines[name] = pipeline state.graph.bl_renderers[name] = BottomLevelRendererState(status='pending', transform=transform, setup_args=setup_args) request_redraw(state) # Let renderer core know it should refresh next frame def add_rasterizer_to_scene_graph(state: WispState, name: str, pipeline: RasterizationPipeline, transform: ObjectTransform = None, **setup_args): """ Adds a new object rasterization pipeline to the scene graph. The object is represented in the SceneGraph by a newly created BottomLevelRenderer. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph pipeline (RasterizationPipeline): A pipeline consisting of a rastrizer. transform (ObjectTransform): The object transform, containing a 4x4 transformation matrix which transforms the object from local object space to world space. setup_args (Dict): Optional setup args which control how the BottomLevelRenderer will be created. """ if transform is None: transform = ObjectTransform(device=pipeline.rasterizer.device) state.graph.neural_pipelines[name] = pipeline state.graph.bl_renderers[name] = BottomLevelRendererState(status='pending', transform=transform, setup_args=setup_args) request_redraw(state) # Let renderer core know it should refresh next frame The provided code snippet includes necessary dependencies for implementing the `add_to_scene_graph` function. Write a Python function `def add_to_scene_graph(state: WispState, name: str, obj, transform: ObjectTransform = None, **setup_args)` to solve the following problem: Adds a new object to the scene graph. obj can be any supported object type, neural or non-neural. This is the most general function used to manage adding new objects to the scene graph. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph obj (object): Any object supported by the scene-graph. i.e: for neural fields, obj is a Pipeline. transform (ObjectTransform): The object transform, containing a 4x4 transformation matrix which transforms the object from local object space to world space. setup_args (Dict): Optional setup args which control how the BottomLevelRenderer will be created. Here is the function: def add_to_scene_graph(state: WispState, name: str, obj, transform: ObjectTransform = None, **setup_args): """ Adds a new object to the scene graph. obj can be any supported object type, neural or non-neural. This is the most general function used to manage adding new objects to the scene graph. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph obj (object): Any object supported by the scene-graph. i.e: for neural fields, obj is a Pipeline. transform (ObjectTransform): The object transform, containing a 4x4 transformation matrix which transforms the object from local object space to world space. setup_args (Dict): Optional setup args which control how the BottomLevelRenderer will be created. """ if isinstance(obj, Pipeline): add_pipeline_to_scene_graph(state, name, obj, transform, **setup_args) elif isinstance(obj, RasterizationPipeline): add_rasterizer_to_scene_graph(state, name, obj, transform, **setup_args) else: # TODO (operel): Currently only the above pipelines are supported raise NotImplementedError(f'Unsupported object type added to scene graph: {obj}')
Adds a new object to the scene graph. obj can be any supported object type, neural or non-neural. This is the most general function used to manage adding new objects to the scene graph. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph obj (object): Any object supported by the scene-graph. i.e: for neural fields, obj is a Pipeline. transform (ObjectTransform): The object transform, containing a 4x4 transformation matrix which transforms the object from local object space to world space. setup_args (Dict): Optional setup args which control how the BottomLevelRenderer will be created.
22,997
import torch from wisp.core import ObjectTransform from wisp.models import Pipeline, RasterizationPipeline from wisp.framework import WispState, BottomLevelRendererState def request_redraw(state): """ Marks the canvas as dirty, forcing the renderer core to refresh the object renderers on the next rendering iteration. """ state.renderer.canvas_dirty = True The provided code snippet includes necessary dependencies for implementing the `remove_from_scene_graph` function. Write a Python function `def remove_from_scene_graph(state: WispState, name: str)` to solve the following problem: Removes an existing pipeline from the scene graph. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph Here is the function: def remove_from_scene_graph(state: WispState, name: str): """ Removes an existing pipeline from the scene graph. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph """ assert name in state.graph.neural_pipelines, f'Scene graph requested to remove non-existing object: {name}' del state.graph.neural_pipelines[name] del state.graph.bl_renderers[name] request_redraw(state) # Let renderer core know it should refresh next frame
Removes an existing pipeline from the scene graph. Args: state (WispState): A wisp state object, containing the scene graph information. name (str): Unique name of object added to the scene graph
22,998
from __future__ import annotations from collections import defaultdict, deque from typing import Type, TYPE_CHECKING, Union from wisp.models import Pipeline, RasterizationPipeline from wisp.models.nefs import BaseNeuralField from wisp.tracers import BaseTracer def _neural_field_to_renderer_cls(pipeline: Pipeline) -> Type[RayTracedRenderer]: tracer_type = type(pipeline.tracer) # Start by iterating the current tracer type - look for renderers compatible with the current nef type # or any of its parents (the hierarchy of nefs take precedence over the hierarchy of tracers). renderer_cls = None while tracer_type: tracer_name = tracer_type.__name__ # Look for a renderer compatible with the current tracer type and nef classes type_queue = deque([type(pipeline.nef)]) renderer_cls = None while type_queue: # Query nef + tracer combo field_type = type_queue.popleft() field_name = field_type.__name__ supported_tracers = _REGISTERED_RENDERABLE_NEURAL_FIELDS.get(field_name) renderer_cls = supported_tracers.get(tracer_name) if supported_tracers is not None else None # Current nef + tracer pair doesn't match any registered renderer if renderer_cls is not None: break else: # Try querying all parent(nef) + tracer combos for compatibility bases = field_type.__bases__ if len(bases) > 0: for b in bases: type_queue.append(b) if renderer_cls is not None: break # Found a renderer class else: # Didn't find a renderer class - repeat the process with the parent class of the tracer tracer_base_types = tracer_type.__bases__ if len(tracer_base_types) > 0: # Does tracer have a single parent class which inherits from BaseTracer? # If so, keep looking tracer_base_types = [base for base in tracer_type.__bases__ if issubclass(base, BaseTracer)] tracer_type = tracer_base_types[0] if len(tracer_base_types) == 1 else None else: # Reached end of tracers hierarchy or it is too ambiguous, quit and fail gracefully tracer_type = None if tracer_type is None: raise ValueError(f'Renderer factory encountered an unknown neural pipeline: ' f'Neural Field {type(pipeline.nef).__name__} with tracer {type(pipeline.tracer).__name__}. ' 'Please register the factory to reflect what kind of renderer should be created for this ' 'type of neural field/tracer.') return renderer_cls def _neural_rasterizer_to_renderer_cls(pipeline: RasterizationPipeline) -> Type[RasterizedRenderer]: # Start by iterating the current rasterizer_type type - # look for renderers compatible with the current rasterizer_type type or any of its parents renderer_cls = None type_queue = deque([type(pipeline.rasterizer)]) while type_queue: # Query rasterizer rast_type = type_queue.popleft() rast_name = rast_type.__name__ # Look for a renderer compatible with the current rasterizer type renderer_cls = _REGISTERED_RENDERABLE_RASTERIZERS.get(rast_name) if renderer_cls is not None: break # Matching renderer found else: # Current rasterizer class doesn't match any registered renderer # Try querying all parent(rast_type) for compatibility bases = rast_type.__bases__ if len(bases) > 0: for b in bases: type_queue.append(b) if renderer_cls is None: raise ValueError(f'Renderer factory encountered an unknown neural rasterization pipeline: ' f'{type(pipeline.rasterizer).__name__}. ' 'Please register the factory to reflect what kind of renderer should be created for this ' 'type of rasterization pipeline.') return renderer_cls class BottomLevelRenderer(WispModule, ABC): def __init__(self, *args, **kwargs): super().__init__() self._data_layers = dict() def pre_render(self, payload: FramePayload, *args, **kwargs) -> None: """ General frame setup occurs here, for example: 1. Ray Tracers - ray generation. 2. Update shader uniforms. """ pass def needs_refresh(self, payload: FramePayload, *args, **kwargs) -> bool: """ Override to optimize cases when the neural field does not require rendering from scratch. """ return True def render(self, *args, **kwargs) -> RenderBuffer: raise NotImplementedError('BottomLevelRenderer subclasses must implement render() logic.') def redraw(self): if self.needs_redraw(): self._data_layers = self.regenerate_data_layers() def post_render(self, *args, **kwargs) -> None: """ General frame teardown logic takes place, for example: 1. Cleanup of temporary information generated during the frame 2. Caching of information relevant for the next frames. """ pass def needs_redraw(self) -> bool: return True def regenerate_data_layers(self) -> Dict[str, PrimitivesPack]: return dict() def data_layers(self) -> Dict[str, PrimitivesPack]: """ Returns layers of information made of primitives, visually describing the renderer internal structures """ return self._data_layers def device(self) -> torch.device: raise NotImplementedError('BottomLevelRenderer subclasses must implement device') def dtype(self) -> torch.dtype: raise NotImplementedError('BottomLevelRenderer subclasses must implement dtype') def acceleration_structure(self) -> Optional[str]: """ Returns a descriptive name of the acceleration structure used by this object, if applicable. """ return None def features_structure(self) -> Optional[str]: """ Returns a descriptive name of the feature structure used by this object, if applicable. """ return None def public_properties(self) -> Dict[str, Any]: """ Wisp modules expose their public properties in a dictionary. The purpose of this method is to give an easy table of outwards facing attributes, for the purpose of logging, gui apps, etc. BLASGrids are generally assumed to contain a bottom level acceleration structure. """ return dict() def create_neural_field_renderer(neural_object: Union[Pipeline, RasterizationPipeline], **kwargs) -> BottomLevelRenderer: # Fetch the neural field object. If given a pipeline, query the nef field. # Otherwise assume we're given the neural field explicitly. if isinstance(neural_object, Pipeline): # Obtain the appropriate renderer class, compatible with this neural field type and tracer renderer_cls = _neural_field_to_renderer_cls(neural_object) # In the case of a pipeline, we use a specialized constructor which builds the renderer from its components # kwargs may override the pipeline default settings renderer_instance = renderer_cls.from_pipeline(neural_object, **kwargs) elif isinstance(neural_object, RasterizationPipeline): renderer_cls = _neural_rasterizer_to_renderer_cls(neural_object) # In the case of a rasterization pipeline, # we use a specialized constructor which builds the renderer from its components # kwargs may override the pipeline default settings renderer_instance = renderer_cls.from_pipeline(neural_object, **kwargs) else: raise NotImplementedError(f"Unknown pipeline type: {neural_object}") return renderer_instance
null
22,999
from __future__ import annotations from typing import Type from wisp.models.nefs import BaseNeuralField from wisp.tracers import BaseTracer from wisp.renderer.core.api.base_renderer import BottomLevelRenderer from wisp.renderer.core.api.renderers_factory import register_neural_field_type, register_rasterizer_type class BottomLevelRenderer(WispModule, ABC): def __init__(self, *args, **kwargs): super().__init__() self._data_layers = dict() def pre_render(self, payload: FramePayload, *args, **kwargs) -> None: """ General frame setup occurs here, for example: 1. Ray Tracers - ray generation. 2. Update shader uniforms. """ pass def needs_refresh(self, payload: FramePayload, *args, **kwargs) -> bool: """ Override to optimize cases when the neural field does not require rendering from scratch. """ return True def render(self, *args, **kwargs) -> RenderBuffer: raise NotImplementedError('BottomLevelRenderer subclasses must implement render() logic.') def redraw(self): if self.needs_redraw(): self._data_layers = self.regenerate_data_layers() def post_render(self, *args, **kwargs) -> None: """ General frame teardown logic takes place, for example: 1. Cleanup of temporary information generated during the frame 2. Caching of information relevant for the next frames. """ pass def needs_redraw(self) -> bool: return True def regenerate_data_layers(self) -> Dict[str, PrimitivesPack]: return dict() def data_layers(self) -> Dict[str, PrimitivesPack]: """ Returns layers of information made of primitives, visually describing the renderer internal structures """ return self._data_layers def device(self) -> torch.device: raise NotImplementedError('BottomLevelRenderer subclasses must implement device') def dtype(self) -> torch.dtype: raise NotImplementedError('BottomLevelRenderer subclasses must implement dtype') def acceleration_structure(self) -> Optional[str]: """ Returns a descriptive name of the acceleration structure used by this object, if applicable. """ return None def features_structure(self) -> Optional[str]: """ Returns a descriptive name of the feature structure used by this object, if applicable. """ return None def public_properties(self) -> Dict[str, Any]: """ Wisp modules expose their public properties in a dictionary. The purpose of this method is to give an easy table of outwards facing attributes, for the purpose of logging, gui apps, etc. BLASGrids are generally assumed to contain a bottom level acceleration structure. """ return dict() def register_neural_field_type(neural_field_type: Type[BaseNeuralField], tracer_type: Type[BaseTracer], renderer_type: Type[BottomLevelRenderer]): """ Register new types of neural fields with their associated bottom level renderers using this function. This allows the interactive renderer to display this neural field type on the canvas. """ field_name = neural_field_type.__name__ tracer_name = tracer_type.__name__ _REGISTERED_RENDERABLE_NEURAL_FIELDS[field_name][tracer_name] = renderer_type The provided code snippet includes necessary dependencies for implementing the `field_renderer` function. Write a Python function `def field_renderer(field_type: Type[BaseNeuralField], tracer_type: Type[BaseTracer])` to solve the following problem: A decorator that registers a neural field type with a renderer. By registering the renderer type, the interactive renderer knows what type of renderer to create when dealing with this type of field. Essentially, this allows displaying custom types of objects on the canvas. Here is the function: def field_renderer(field_type: Type[BaseNeuralField], tracer_type: Type[BaseTracer]): """ A decorator that registers a neural field type with a renderer. By registering the renderer type, the interactive renderer knows what type of renderer to create when dealing with this type of field. Essentially, this allows displaying custom types of objects on the canvas. """ def _register_renderer_fn(renderer_class: Type[BottomLevelRenderer]): register_neural_field_type(field_type, tracer_type, renderer_class) return renderer_class return _register_renderer_fn
A decorator that registers a neural field type with a renderer. By registering the renderer type, the interactive renderer knows what type of renderer to create when dealing with this type of field. Essentially, this allows displaying custom types of objects on the canvas.
23,000
from __future__ import annotations from typing import Type from wisp.models.nefs import BaseNeuralField from wisp.tracers import BaseTracer from wisp.renderer.core.api.base_renderer import BottomLevelRenderer from wisp.renderer.core.api.renderers_factory import register_neural_field_type, register_rasterizer_type class BottomLevelRenderer(WispModule, ABC): def __init__(self, *args, **kwargs): super().__init__() self._data_layers = dict() def pre_render(self, payload: FramePayload, *args, **kwargs) -> None: """ General frame setup occurs here, for example: 1. Ray Tracers - ray generation. 2. Update shader uniforms. """ pass def needs_refresh(self, payload: FramePayload, *args, **kwargs) -> bool: """ Override to optimize cases when the neural field does not require rendering from scratch. """ return True def render(self, *args, **kwargs) -> RenderBuffer: raise NotImplementedError('BottomLevelRenderer subclasses must implement render() logic.') def redraw(self): if self.needs_redraw(): self._data_layers = self.regenerate_data_layers() def post_render(self, *args, **kwargs) -> None: """ General frame teardown logic takes place, for example: 1. Cleanup of temporary information generated during the frame 2. Caching of information relevant for the next frames. """ pass def needs_redraw(self) -> bool: return True def regenerate_data_layers(self) -> Dict[str, PrimitivesPack]: return dict() def data_layers(self) -> Dict[str, PrimitivesPack]: """ Returns layers of information made of primitives, visually describing the renderer internal structures """ return self._data_layers def device(self) -> torch.device: raise NotImplementedError('BottomLevelRenderer subclasses must implement device') def dtype(self) -> torch.dtype: raise NotImplementedError('BottomLevelRenderer subclasses must implement dtype') def acceleration_structure(self) -> Optional[str]: """ Returns a descriptive name of the acceleration structure used by this object, if applicable. """ return None def features_structure(self) -> Optional[str]: """ Returns a descriptive name of the feature structure used by this object, if applicable. """ return None def public_properties(self) -> Dict[str, Any]: """ Wisp modules expose their public properties in a dictionary. The purpose of this method is to give an easy table of outwards facing attributes, for the purpose of logging, gui apps, etc. BLASGrids are generally assumed to contain a bottom level acceleration structure. """ return dict() def register_rasterizer_type(rasterizer_type: Type, renderer_type: Type[BottomLevelRenderer]): """ Register new types of rasterizers with their associated bottom level renderers using this function. This allows the interactive renderer to display this rasterizer type on the canvas. """ _REGISTERED_RENDERABLE_RASTERIZERS[rasterizer_type.__name__] = renderer_type The provided code snippet includes necessary dependencies for implementing the `register_rasterizer` function. Write a Python function `def register_rasterizer(rasterizer_type: Type)` to solve the following problem: A decorator that registers a rasterizer type with a renderer. By registering the renderer type, the interactive renderer knows what type of renderer to create when dealing with this type of rasterizer. Essentially, this allows displaying custom types of objects on the canvas. Here is the function: def register_rasterizer(rasterizer_type: Type): """ A decorator that registers a rasterizer type with a renderer. By registering the renderer type, the interactive renderer knows what type of renderer to create when dealing with this type of rasterizer. Essentially, this allows displaying custom types of objects on the canvas. """ def _register_renderer_fn(renderer_class: Type[BottomLevelRenderer]): register_rasterizer_type(rasterizer_type, renderer_class) return renderer_class return _register_renderer_fn
A decorator that registers a rasterizer type with a renderer. By registering the renderer type, the interactive renderer knows what type of renderer to create when dealing with this type of rasterizer. Essentially, this allows displaying custom types of objects on the canvas.
23,001
from __future__ import annotations import copy import torch import wisp.framework.state as state from wisp.renderer.core.control.camera_controller_mode import CameraControlMode from wisp.renderer.core.control.io import WispMouseButton def quat_mul(Q1, Q2): return torch.tensor([Q1[0] * Q2[3] + Q1[3] * Q2[0] - Q1[2] * Q2[1] + Q1[1] * Q2[2], Q1[1] * Q2[3] + Q1[2] * Q2[0] + Q1[3] * Q2[1] - Q1[0] * Q2[2], Q1[2] * Q2[3] - Q1[1] * Q2[0] + Q1[0] * Q2[1] + Q1[3] * Q2[2], Q1[3] * Q2[3] - Q1[0] * Q2[0] - Q1[1] * Q2[1] - Q1[2] * Q2[2]])
null
23,002
from __future__ import annotations import copy import torch import wisp.framework.state as state from wisp.renderer.core.control.camera_controller_mode import CameraControlMode from wisp.renderer.core.control.io import WispMouseButton def quat_matrix(q): # True only for unit quaternions xx = q[0] * q[0] xy = q[0] * q[1] xz = q[0] * q[2] xw = q[0] * q[3] yy = q[1] * q[1] yz = q[1] * q[2] yw = q[1] * q[3] zz = q[2] * q[2] zw = q[2] * q[3] ww = q[3] * q[3] return torch.tensor([[ww + xx - yy - zz, 2.0 * (xy - zw), 2.0 * (xz + yw), 0.0], [2.0 * (xy + zw), ww - xx + yy - zz, 2.0 * (yz - xw), 0.0], [2.0 * (xz - yw), 2.0 * (yz + xw), ww - xx - yy + zz, 0.0], [0.0, 0.0, 0.0, 1.0]], dtype=torch.float64)
null
23,003
from __future__ import annotations import abc import math import numpy as np import torch import copy from collections import defaultdict from typing import Dict, List, Iterable, Tuple from kaolin.render.camera import Camera, PinholeIntrinsics, OrthographicIntrinsics from wisp.framework import WispState, BottomLevelRendererState from wisp.core import RenderBuffer, Rays, PrimitivesPack, create_default_channel, ObjectTransform from wisp.ops.raygen import generate_pinhole_rays, generate_ortho_rays, generate_centered_pixel_coords from wisp.renderer.core.api import BottomLevelRenderer, RayTracedRenderer, FramePayload, create_neural_field_renderer from wisp.gfx.datalayers import CameraDatalayers class RendererCore: def __init__(self, state: WispState): self.state = state self.device = state.renderer.device # Create a camera for user view self.camera = self._setup_camera(state) self._camera_layers = CameraDatalayers() # Set up the list of available bottom level object renderers, according to scene graph self._renderers = None self._tlas = None self.refresh_bl_renderers(state) self.res_x, self.res_y = None, None self.set_full_resolution() self._last_state = dict() self._last_renderbuffer = None # Minimal resolution supported by RendererCore self.MIN_RES = 64 def _default_camera(self, lens="perspective"): # TODO: move defaults elsewhere if lens == 'perspective': return Camera.from_args( eye=torch.tensor([6.0, 6.0, 6.0]), at=torch.tensor([0.0, 0.0, 0.0]), up=torch.tensor([0.0, 1.0, 0.0]), fov=40 * np.pi / 180, # In radians x0=0.0, y0=0.0, width=900, height=675, near=1e-2, far=30, dtype=torch.float64, device=self.device ) elif lens == 'orthographic': return Camera.from_args( eye=torch.tensor([6.0, 6.0, 6.0]), at=torch.tensor([0.0, 0.0, 0.0]), up=torch.tensor([0.0, 1.0, 0.0]), width=800, height=800, near=-800, far=800, fov_distance=1.0, dtype=torch.float64, device=self.device ) def change_camera_projection_mode(self, lens: str) -> None: # Need to update only the intrinsic component of the camera # TODO (operel): in the future, kaolin should implement a capability to switch lens and approximate params # from one type to another where applicable if lens == 'Perspective': intrinsics = PinholeIntrinsics.from_fov( fov=30 * np.pi / 180, # In radians x0=0.0, y0=0.0, width=self.camera.width, height=self.camera.height, near=1e-2, far=1e2, dtype=self.camera.dtype, device=self.camera.device ) elif lens == 'Orthographic': intrinsics = OrthographicIntrinsics.from_frustum( width=self.camera.width, height=self.camera.height, near=-self.camera.width, far=self.camera.height, fov_distance=1.0, dtype=self.camera.dtype, device=self.camera.device ) else: raise ValueError(f"Unknown lens type: {lens} given to render_core.change_camera_projection_mode") self.camera.intrinsics = intrinsics del self._last_state['camera'] # The previous camera is now obsolete def _setup_camera(self, state: WispState): # Use selected camera to control canvas camera = state.renderer.selected_camera if camera is None: # Create a default camera lens_type = self.state.renderer.selected_camera_lens camera = self._default_camera(lens_type) camera = camera.to(self.device) return camera def refresh_bl_renderers(self, state: WispState) -> None: renderers = dict() scene_graph = state.graph # Remove obsolete bottom level renderers for pipelines that no longer exist for obj_name in list(scene_graph.bl_renderers.keys()): if obj_name not in scene_graph.neural_pipelines: del scene_graph.bl_renderers[obj_name] # Set up a renderer for all neural pipelines in the scene for renderer_id, neural_pipeline in scene_graph.neural_pipelines.items(): # See if a descriptor for the renderer already exists. bl_state = scene_graph.bl_renderers.get(renderer_id) if bl_state is None: # If not, create a default one object_transform = ObjectTransform(device=self.device,dtype=self.camera.dtype) bl_state = BottomLevelRendererState(status='pending', setup_args=dict(), transform=object_transform) scene_graph.bl_renderers[renderer_id] = bl_state if bl_state.status == 'loaded': assert bl_state.renderer is not None, \ f'status of renderer {renderer_id} shows it was loaded, but renderer instance is None.' renderers[renderer_id] = bl_state.renderer elif bl_state.status == 'pending': bl_state.renderer = create_neural_field_renderer(neural_object=neural_pipeline, **bl_state.setup_args) bl_state.status = 'loaded' renderers[renderer_id] = bl_state.renderer scene_graph.visible_objects[renderer_id] = True # Mark as visible when first loaded elif bl_state.status == 'ignored': pass # Skip renderer on purpose else: raise ValueError(f'Invalid bottom level renderer state: {bl_state.status}') self._renderers = renderers # Refresh TLAS self._tlas = self._setup_tlas(state) def _setup_tlas(self, state: WispState): # Currently the top-level acceleration structure uses a straightforward ordered list stub return ListTLAS(state, self._renderers) def set_full_resolution(self): self.res_x = self.camera.width self.res_y = self.camera.height self.interactive_mode = False def set_low_resolution(self, downscale_factor: int = 4): self.res_x = self.camera.width // downscale_factor self.res_y = self.camera.height // downscale_factor self.interactive_mode = True def resize_canvas(self, width, height): self.camera.intrinsics.width = width self.camera.intrinsics.height = height self.set_low_resolution() def _update_scene_graph(self): """ Update scene graph information about objects and their data layers """ # New data layers maybe have been generated, update the scene graph about their existence. # Some existing layers may have been regenerated, in which case we copy their previous "toggled on/off" status. data_layers = self._bl_renderer_data_layers() for object_id, bl_renderer_state in self.state.graph.bl_renderers.items(): object_layers = data_layers[object_id] bl_renderer_state.data_layers = object_layers toggled_data_layers = defaultdict(bool) for layer_name, layer in object_layers.items(): # Copy over previous toggles if such exist toggled_data_layers[layer_name] = bl_renderer_state.toggled_data_layers.get(layer_name, False) bl_renderer_state.toggled_data_layers = toggled_data_layers def redraw(self) -> None: """ Allow bottom-level renderers to refresh internal information, such as data layers. """ # Read phase: sync with scene graph, create renderers for new objects added self.refresh_bl_renderers(self.state) # Invoke internal redraw() on all current visible renderers, to imply it's time to refresh content # (i.e. data layers may get generate here behind the scenes) scene_graph = self.state.graph for obj_id, renderer in self._renderers.items(): if obj_id in scene_graph.visible_objects: renderer.redraw() # Write phase: update scene graph back with latest info from render core, i.e. new data layers generated self._update_scene_graph() def render(self, time_delta=None, force_render=False) -> RenderBuffer: """Render a frame. Args: time_delta (float): The time delta from the previous frame, used to control renderer parameters based on the amount of detected lag. force_render (bool): If True, will always output a fresh new RenderBuffer. Otherwise the RenderBuffer can be a stale copy of the the previous frame if no updates are detected. Returns: (wisp.core.RenderBuffer): The rendered buffer. """ payload = self._prepare_payload(time_delta) rb = self._render_payload(payload, force_render) output_rb = self._post_render(payload, rb) return output_rb def _prepare_payload(self, time_delta=None) -> FramePayload: """This function will prepare the FramePayload for the current frame. The FramePayload contains metadata for the current frame, from which the RenderBuffer will be generated from. Args: time_delta (float): The time delta from the previous frame, used to control renderer parameters based on the amount of detected lag. Returns: (wisp.renderer.core.api.FramePayload): The metadata for the frame. """ # Adjust resolution of all renderers to maintain FPS camera = self.camera clear_color = self.state.renderer.clear_color_value res_x, res_y = self.res_x, self.res_y # If the FPS is slow, downscale the resolution for the render. if self.interactive_mode and time_delta is not None: #target_delta = 1.0 / self.target_interactive_fps target_delta = 1.0 / 20.0 if 'res_x' in self._last_state: num_pixels = self._last_state['res_x'] * self._last_state['res_y'] else: num_pixels = res_x * res_y time_per_pixel = time_delta / float(num_pixels) target_num_pixels = target_delta / time_per_pixel screen_ratio = res_x / res_y res = math.sqrt(target_num_pixels / screen_ratio) res_x = min(res_x, int(math.floor(res * screen_ratio))) res_y = min(res_y, int(math.floor(res))) if res_y < self.MIN_RES: res_x = int(math.floor(self.MIN_RES*screen_ratio)) res_y = int(math.floor(self.MIN_RES)) if 'res_x' in self._last_state: if abs(res_x - self._last_state['res_x']) < 10: res_x = self._last_state['res_x'] res_y = self._last_state['res_y'] # TODO(ttakikawa): Leaving a note here to think about whether this should be the case... # The renderer always needs depth, alpha, and rgb required_channels = {"rgb", "depth", "alpha"} selected_canvas_channel = self.state.renderer.selected_canvas_channel.lower() visible_objects = set([k for k,v in self.state.graph.visible_objects.items() if v]) payload = FramePayload(camera=camera, interactive_mode=self.interactive_mode, render_res_x=res_x, render_res_y=res_y, time_delta=time_delta, visible_objects=visible_objects, clear_color=clear_color, channels={selected_canvas_channel}.union(required_channels)) for renderer_id, renderer in self._renderers.items(): if renderer_id in payload.visible_objects: renderer.pre_render(payload) return payload def raygen(self, camera, res_x, res_y): ray_grid = generate_centered_pixel_coords(camera.width, camera.height, res_x, res_y, device=self.device) if camera.lens_type == 'pinhole': rays = generate_pinhole_rays(camera, ray_grid) elif camera.lens_type == 'ortho': rays = generate_ortho_rays(camera, ray_grid) else: raise ValueError(f'RendererCore failed to raygen on unknown camera lens type: {camera.lens_type}') return rays def _create_empty_rb(self, height, width, dtype=torch.float32) -> RenderBuffer: clear_color = self.state.renderer.clear_color_value clear_depth = self.state.renderer.clear_depth_value return RenderBuffer( rgb=torch.tensor(clear_color, dtype=dtype, device=self.device).repeat(height, width, 1), alpha=torch.zeros((height, width, 1), dtype=dtype, device=self.device), depth=torch.full((height, width, 1), fill_value=clear_depth, dtype=dtype, device=self.device), hit=None ) def _render_payload(self, payload: FramePayload, force_render: bool) -> RenderBuffer: """Renders a RenderBuffer using a FramePayload which contains metadata. Args: payload (wisp.renderer.core.api.FramePayload): Metadata for the frame to be renderered. force_render (bool): If True, will always output a fresh new RenderBuffer. Otherwise the RenderBuffer can be a stale copy of the the previous frame if no updates are detected. Returns: (wisp.core.RenderBuffer): The rendered buffer. """ camera = payload.camera res_x, res_y = payload.render_res_x, payload.render_res_y visible_renderers = [r for r_id, r in self._renderers.items() if r_id in payload.visible_objects] renderers_to_refresh = list(filter(lambda renderer: renderer.needs_refresh(payload), visible_renderers)) if not self.needs_refresh() and len(renderers_to_refresh) == 0 and not force_render: return self._last_renderbuffer # No need to regenerate.. # Generate rays rays = self.raygen(camera, res_x, res_y) renderers_in_view = self._tlas.traverse(rays, payload) rb_dtype = torch.float32 clear_depth = self.state.renderer.clear_depth_value out_rb = self._create_empty_rb(height=camera.height, width=camera.width, dtype=rb_dtype) for renderer, hit_rays in renderers_in_view: if isinstance(renderer, RayTracedRenderer): in_rays = hit_rays.to(device=renderer.device, dtype=renderer.dtype) rb = renderer.render(in_rays) else: # RasterizedRenderer # TODO (operel): Handle transformed rasterized objects in_cam = self.camera.to(device=renderer.device, dtype=renderer.dtype) rb = renderer.render(in_cam) rb = rb.to(device=self.device) rb.rgb = rb.rgb.to(dtype=rb_dtype) rb.alpha = rb.alpha.to(dtype=rb_dtype) rb.depth = rb.depth.to(dtype=rb_dtype) # TODO (operel): if rb.depth is None -> painters algorithm # Normalize ray-traced depth buffer to graphics api range img_dims = rb.depth.shape # Clip depth values which fall outside of the view frustum clip_mask = camera.clip_mask(rb.depth.squeeze(-1)) rb.alpha[~clip_mask] = 0.0 # Normalize depth from [0, inf] to NDC space according to camera settings # (depends on near / far and NDC space) ndc_depth = camera.normalize_depth(rb.depth.reshape(-1, 1)) rb.depth = ndc_depth.reshape(img_dims) # Set depth of missed rays to far clipping plane, as PackedRFTracer initializes depth to 0 and writes values # only for hit rays alpha_mask = ~rb.hit[...,0] rb.depth[alpha_mask] = clear_depth rb.depth = rb.depth.to(rb_dtype) out_rb = out_rb.blend(rb, channel_kit=self.state.graph.channels) return out_rb def _post_render(self, payload: FramePayload, rb: RenderBuffer) -> RenderBuffer: # Update current resolution in case it was decreased to maintain fps self.res_x, self.res_y = payload.render_res_x, payload.render_res_y # Cache information to accelerate next frames self._last_renderbuffer = rb # Record last state, to, i.e, calculate if needs to redraw future frames self._last_state['camera'] = copy.deepcopy(payload.camera) self._last_state['res_x'] = payload.render_res_x self._last_state['res_y'] = payload.render_res_y for renderer_id, renderer in self._renderers.items(): if renderer_id in payload.visible_objects: renderer.post_render() # Create an output renderbuffer to contain the currently viewed mode as rgba channel output_rb = self.map_output_channels_to_rgba(rb) return output_rb def needs_refresh(self) -> bool: if len(self._last_state) == 0: return True # Resolution check: if not full resolution - canvas is dirty if self._last_state['res_x'] < self.camera.width or self._last_state['res_y'] < self.camera.height: return True for att_name, prev_val in self._last_state.items(): if not hasattr(self, att_name): continue curr_val = self.__getattribute__(att_name) if isinstance(curr_val, Camera): if not torch.allclose(curr_val, prev_val): return True elif curr_val != prev_val: return True return False def _bl_renderer_data_layers(self) -> Dict[str, PrimitivesPack]: """ Returns the bottom level object data layers""" layers = dict() for renderer_id, renderer in self._renderers.items(): layers[renderer_id] = renderer.data_layers() return layers def _cameras_data_layers(self) -> Iterable[PrimitivesPack]: """ Returns the available cameras data layer (all visible cameras layers are packed together) """ cameras_to_redraw = {camera_id: camera for camera_id, camera in self.state.graph.cameras.items() if self.state.graph.visible_objects.get(camera_id, False)} layers = self._camera_layers.regenerate_data_layers(cameras_to_redraw, self.state.renderer.clear_color_value) return layers.values() def active_data_layers(self) -> List[PrimitivesPack]: layers_to_draw = [] for obj_state in self.state.graph.bl_renderers.values(): for layer_id, layer in obj_state.data_layers.items(): if obj_state.toggled_data_layers[layer_id]: # Attach object transform reference to data layers, assumes the object maintains this transform ref layer.transform = obj_state.transform layers_to_draw.append(layer) camera_data_layers = self._cameras_data_layers() layers_to_draw.extend(camera_data_layers) return layers_to_draw def map_output_channels_to_rgba(self, rb: RenderBuffer): selected_output_channel = self.state.renderer.selected_canvas_channel.lower() rb_channel = rb.get_channel(selected_output_channel) if rb_channel is None: # Unknown channel type configured to view over the canvas. # That can happen if, i.e. no object have traced a RenderBuffer with this channel. # Instead of failing, create an empty rb height, width = rb.rgb.shape[:2] return self._create_empty_rb(height=height, width=width, dtype=rb.rgb.dtype) # Normalize channel to [0, 1] channels_kit = self.state.graph.channels channel_info = channels_kit.get(selected_output_channel, create_default_channel()) normalized_channel = channel_info.normalize_fn(rb_channel.clone()) # Clone to protect from modifications # To RGB (in normalized space) # TODO (operel): incorporate color maps channel_dim = normalized_channel.shape[-1] if channel_dim == 1: rgb = torch.cat((normalized_channel, normalized_channel, normalized_channel), dim=-1) elif channel_dim == 2: rgb = torch.cat((normalized_channel, normalized_channel, torch.zeros_like(normalized_channel)), dim=-1) elif channel_dim == 3: rgb = normalized_channel else: raise ValueError('Cannot display channels with more than 3 dimensions over the canvas.') canvas_rb = RenderBuffer(rgb=rgb, depth=rb.depth, alpha=rb.alpha) return canvas_rb def renderers(self) -> Dict[str, BottomLevelRenderer]: """ All loaded bottom level renderers currently employed by the renderer core """ return self._renderers def camera(self) -> Camera: return self.state.renderer.selected_camera def camera(self, camera: Camera) -> None: self.state.renderer.selected_camera = camera def target_fps(self) -> float: return self.state.renderer.target_fps def target_interactive_fps(self) -> float: return self.state.renderer.target_interactive_fps def target_static_fps(self) -> float: return self.state.renderer.target_static_fps def interactive_mode(self) -> bool: return self.state.renderer.interactive_mode def interactive_mode(self, mode: bool) -> None: self.state.renderer.interactive_mode = mode def selected_camera_lens(self) -> str: return self.state.renderer.selected_camera_lens def selected_camera_lens(self, lens: str): self.state.renderer.selected_camera_lens = lens The provided code snippet includes necessary dependencies for implementing the `enable_amp` function. Write a Python function `def enable_amp(func)` to solve the following problem: An extension to @torch.cuda.amp.autocast which queries WispState to check if mixed precision should be enabled. Here is the function: def enable_amp(func): """ An extension to @torch.cuda.amp.autocast which queries WispState to check if mixed precision should be enabled. """ def _enable_amp(self: RendererCore, *args, **kwargs): with torch.cuda.amp.autocast(enabled=self.state.renderer.enable_amp): return func(self, *args, **kwargs) return _enable_amp
An extension to @torch.cuda.amp.autocast which queries WispState to check if mixed precision should be enabled.
23,004
import os from contextlib import contextmanager if not os.environ.get('ENABLE_PYCUDA') == '1': from cuda import cuda import torch def cuda_map_resource(img): """Context manager simplifying use of cuda.cuGraphicsMapResources / cuGraphicsSubResourceGetMappedArray. Boilerplate code based in part on pytorch-glumpy. """ # args: (count, resource, stream) mapping_result = cuda.cuGraphicsMapResources(1, img, torch.cuda.default_stream().cuda_stream) if mapping_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to map GL graphics resource to be accessed by CUDA.") # args (resource, arrayIndex, mipLevel) mapping_array = cuda.cuGraphicsSubResourceGetMappedArray(img, 0, 0) if mapping_array[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to get mapped array from GL resource for CUDA copy.") yield mapping_array[1] unmapping_result = cuda.cuGraphicsUnmapResources(1, img, torch.cuda.default_stream().cuda_stream) if unmapping_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to unmap GL graphics resource from being accessed by CUDA.") else: import pycuda import pycuda.gl as pycuda_gl window = None try: # !!! Should be called when interactive wisp loads, before any torch ops take place !!! # The following is a hacky workaround due to a cublas error on interfering streams: # pycuda.gl fails to initialize after torch performs batched matrix multiplication # (the bugs causes any following torch.dot invocations to fail).. # The solution is to initialize pycuda.gl early when wisp loads. # To load pycuda.gl, we require a GL context of some window, # so here we force glumpy-glfw to create an invisible window, which generates an opengl context. # Then immediately import pycuda.gl.autoint to let it initialize properly from glumpy import app # Tell glumpy to use glfw backend app.use("glfw_imgui") # Let glumpy use glfw to create an invisible window window = app.Window(width=10, height=10, title='dummy', visible=False) # pycuda initializes the default context with "cuGLCtxCreate", but this call will fail if a GL context # is not currently set. Therefore import is invoked only after glfw obtains a GL context. # See: https://documen.tician.de/pycuda/gl.html#module-pycuda.gl.autoinit import pycuda.gl.autoinit # Next tell torch to initialize the primary cuda context import torch torch.cuda.init() # pycuda should not create a new context, but retain the torch one import pycuda.driver as cuda pycuda_context = cuda.Device(0).retain_primary_context() except (ModuleNotFoundError, ImportError): pass # Don't fail if interactive mode is disabled (e.g: glumpy or pycuda are unavailable) finally: if window is not None: window.close() def cuda_map_resource(img): """Context manager simplifying use of pycuda_gl.RegisteredImage. Boilerplate code based in part on pytorch-glumpy. """ mapping = img.map() yield mapping.array(0, 0) mapping.unmap() def cuda_2d_memcpy(resource_handle, shared_tex, img, height): cpy = cuda.CUDA_MEMCPY2D() with cuda_map_resource(resource_handle) as ary: cpy.srcDevice = img.data_ptr() cpy.srcMemoryType = cuda.CUmemorytype.CU_MEMORYTYPE_DEVICE cpy.dstArray = ary cpy.dstMemoryType = cuda.CUmemorytype.CU_MEMORYTYPE_ARRAY cpy.WidthInBytes = cpy.srcPitch = cpy.dstPitch = shared_tex.nbytes // height cpy.Height = height cpy_result = cuda.cuMemcpy2DUnaligned(cpy) if cpy_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to memcopy cuda buffer to GL.")
null
23,005
import os from contextlib import contextmanager def cuda_register_gl_image(image, target): # Create shared GL / CUDA resource map_flags = cuda.CUgraphicsRegisterFlags.CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD register_result = cuda.cuGraphicsGLRegisterImage(image=image, target=target, Flags=map_flags) if register_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError('Failed to register GL texture as a CUDA shared resource.') resource_handle = register_result[1] return resource_handle
null
23,006
import os from contextlib import contextmanager def cuda_unregister_resource(handle): unregister_result = cuda.cuGraphicsUnregisterResource(handle) if unregister_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError('Failed to unregister CUDA resource.')
null
23,007
import os from contextlib import contextmanager if not os.environ.get('ENABLE_PYCUDA') == '1': from cuda import cuda import torch def cuda_map_resource(img): """Context manager simplifying use of cuda.cuGraphicsMapResources / cuGraphicsSubResourceGetMappedArray. Boilerplate code based in part on pytorch-glumpy. """ # args: (count, resource, stream) mapping_result = cuda.cuGraphicsMapResources(1, img, torch.cuda.default_stream().cuda_stream) if mapping_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to map GL graphics resource to be accessed by CUDA.") # args (resource, arrayIndex, mipLevel) mapping_array = cuda.cuGraphicsSubResourceGetMappedArray(img, 0, 0) if mapping_array[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to get mapped array from GL resource for CUDA copy.") yield mapping_array[1] unmapping_result = cuda.cuGraphicsUnmapResources(1, img, torch.cuda.default_stream().cuda_stream) if unmapping_result[0] != cuda.CUresult.CUDA_SUCCESS: raise RuntimeError("Failed to unmap GL graphics resource from being accessed by CUDA.") else: import pycuda import pycuda.gl as pycuda_gl window = None try: # !!! Should be called when interactive wisp loads, before any torch ops take place !!! # The following is a hacky workaround due to a cublas error on interfering streams: # pycuda.gl fails to initialize after torch performs batched matrix multiplication # (the bugs causes any following torch.dot invocations to fail).. # The solution is to initialize pycuda.gl early when wisp loads. # To load pycuda.gl, we require a GL context of some window, # so here we force glumpy-glfw to create an invisible window, which generates an opengl context. # Then immediately import pycuda.gl.autoint to let it initialize properly from glumpy import app # Tell glumpy to use glfw backend app.use("glfw_imgui") # Let glumpy use glfw to create an invisible window window = app.Window(width=10, height=10, title='dummy', visible=False) # pycuda initializes the default context with "cuGLCtxCreate", but this call will fail if a GL context # is not currently set. Therefore import is invoked only after glfw obtains a GL context. # See: https://documen.tician.de/pycuda/gl.html#module-pycuda.gl.autoinit import pycuda.gl.autoinit # Next tell torch to initialize the primary cuda context import torch torch.cuda.init() # pycuda should not create a new context, but retain the torch one import pycuda.driver as cuda pycuda_context = cuda.Device(0).retain_primary_context() except (ModuleNotFoundError, ImportError): pass # Don't fail if interactive mode is disabled (e.g: glumpy or pycuda are unavailable) finally: if window is not None: window.close() def cuda_map_resource(img): """Context manager simplifying use of pycuda_gl.RegisteredImage. Boilerplate code based in part on pytorch-glumpy. """ mapping = img.map() yield mapping.array(0, 0) mapping.unmap() def cuda_2d_memcpy(resource_handle, shared_tex, img, height): cpy = pycuda.driver.Memcpy2D() with cuda_map_resource(resource_handle) as ary: cpy.set_src_device(img.data_ptr()) cpy.set_dst_array(ary) cpy.width_in_bytes = cpy.src_pitch = cpy.dst_pitch = shared_tex.nbytes // height cpy.height = height cpy(aligned=False)
null
23,008
import os from contextlib import contextmanager def cuda_register_gl_image(image, target): # Create shared GL / CUDA resource map_flags = pycuda_gl.graphics_map_flags.WRITE_DISCARD resource_handle = pycuda_gl.RegisteredImage(image, target, map_flags) return resource_handle
null
23,009
import os from contextlib import contextmanager def cuda_unregister_resource(handle): # Nothing to be done - when ref count reaches zero on proxy object in python, unregister should # be called automatically pass
null
23,010
from __future__ import annotations import contextlib import glob import io import logging import math import os import queue import PIL.Image import re import threading import time import torch import torchvision from typing import Literal from wisp.framework import WispState from wisp.renderer.core import RendererCore from wisp.renderer.core.control.io import WispMouseButton from wisp.renderer.core.control.turntable import TurntableCameraMode def hold_canvas(*args, **kwargs): raise RuntimeError(__IPYCANVAS_ERR)
null
23,011
from __future__ import annotations import contextlib import glob import io import logging import math import os import queue import PIL.Image import re import threading import time import torch import torchvision from typing import Literal from wisp.framework import WispState from wisp.renderer.core import RendererCore from wisp.renderer.core.control.io import WispMouseButton from wisp.renderer.core.control.turntable import TurntableCameraMode def dummy_ctx_manager(): yield None
null
23,012
from __future__ import annotations import contextlib import glob import io import logging import math import os import queue import PIL.Image import re import threading import time import torch import torchvision from typing import Literal from wisp.framework import WispState from wisp.renderer.core import RendererCore from wisp.renderer.core.control.io import WispMouseButton from wisp.renderer.core.control.turntable import TurntableCameraMode The provided code snippet includes necessary dependencies for implementing the `make_render_closure` function. Write a Python function `def make_render_closure(render_core: RendererCore, downscale_factor: int = 4)` to solve the following problem: Makes a render closure over input args, so render can be called without arguments. Args: render_core: the RendererCore to use for rendering downscale_factor: how much to downscale the image when rendering Returns: function() -> torch.Tensor 0..1 float, 4 x H x W, where H, W are determined by render_core.camera and the downscale_factor Here is the function: def make_render_closure(render_core: RendererCore, downscale_factor: int = 4): """Makes a render closure over input args, so render can be called without arguments. Args: render_core: the RendererCore to use for rendering downscale_factor: how much to downscale the image when rendering Returns: function() -> torch.Tensor 0..1 float, 4 x H x W, where H, W are determined by render_core.camera and the downscale_factor """ def _render(td): """ Returns 0..1 float torch tensor of size 4 x H x W """ rescale = None if downscale_factor == 1: render_core.set_full_resolution() else: render_core.set_low_resolution(downscale_factor) # TODO(operel): this should not be necessary rescale = torchvision.transforms.Resize((render_core.res_y, render_core.res_x)) renderbuffer = render_core.render(time_delta=td) # we only request rerender when dirty res = renderbuffer.image().rgba.permute(2, 0, 1) / 255 # C x H x W 0..1 if rescale is not None: res = rescale(res) return res return _render
Makes a render closure over input args, so render can be called without arguments. Args: render_core: the RendererCore to use for rendering downscale_factor: how much to downscale the image when rendering Returns: function() -> torch.Tensor 0..1 float, 4 x H x W, where H, W are determined by render_core.camera and the downscale_factor
23,013
from __future__ import annotations import contextlib import glob import io import logging import math import os import queue import PIL.Image import re import threading import time import torch import torchvision from typing import Literal from wisp.framework import WispState from wisp.renderer.core import RendererCore from wisp.renderer.core.control.io import WispMouseButton from wisp.renderer.core.control.turntable import TurntableCameraMode WISP_ROOT_DIR = os.path.realpath( os.path.join(os.path.dirname(os.path.realpath(__file__)), os.path.pardir, os.path.pardir, os.path.pardir)) The provided code snippet includes necessary dependencies for implementing the `save_canvas_render` function. Write a Python function `def save_canvas_render(canvas, filename, save_dir=None)` to solve the following problem: Convenience function to save a rendered frame to a default location, while appending a counter to the basename. Args: canvas: IpyCanvas canvas object filename: filename and extension where to save, note number will be appended `frame.png` --> `frame1.png` save_dir: directory where to save file; will use default _results/jupyter_renders if not provided Here is the function: def save_canvas_render(canvas, filename, save_dir=None): """ Convenience function to save a rendered frame to a default location, while appending a counter to the basename. Args: canvas: IpyCanvas canvas object filename: filename and extension where to save, note number will be appended `frame.png` --> `frame1.png` save_dir: directory where to save file; will use default _results/jupyter_renders if not provided """ if save_dir is None: save_dir = os.path.join(WISP_ROOT_DIR, '_results', 'jupyter_renders') os.makedirs(save_dir, exist_ok=True) basename, extension = os.path.splitext(os.path.basename(filename)) pattern = re.compile(r"%s(\d+)%s" % (basename, extension)) fnames = [ os.path.basename(x) for x in glob.glob(os.path.join(save_dir, "%s[0-9]*%s" % (basename, extension))) ] counts = [int(m.groups()[0]) for m in [pattern.match(f) for f in fnames] if m is not None] count = 0 if len(counts) > 0: count = max(counts) + 1 filename = os.path.join(save_dir, "%s%02d%s" % (basename, count, extension)) logging.info(f'Saving frame to: {filename}') canvas.to_file(filename)
Convenience function to save a rendered frame to a default location, while appending a counter to the basename. Args: canvas: IpyCanvas canvas object filename: filename and extension where to save, note number will be appended `frame.png` --> `frame1.png` save_dir: directory where to save file; will use default _results/jupyter_renders if not provided
23,014
from __future__ import annotations import contextlib import glob import io import logging import math import os import queue import PIL.Image import re import threading import time import torch import torchvision from typing import Literal from wisp.framework import WispState from wisp.renderer.core import RendererCore from wisp.renderer.core.control.io import WispMouseButton from wisp.renderer.core.control.turntable import TurntableCameraMode The provided code snippet includes necessary dependencies for implementing the `np_img_to_compressed_bytes` function. Write a Python function `def np_img_to_compressed_bytes(np_img, format)` to solve the following problem: Converts numpy array to bytes in the specified image format. Args: np_img: numpy array H x W x C uint8 format: any format supported by Pillow, e.g. 'png' or 'jpeg' (note jpeg does not accept RGBA) Return: bytes Here is the function: def np_img_to_compressed_bytes(np_img, format): """ Converts numpy array to bytes in the specified image format. Args: np_img: numpy array H x W x C uint8 format: any format supported by Pillow, e.g. 'png' or 'jpeg' (note jpeg does not accept RGBA) Return: bytes """ img = PIL.Image.fromarray(np_img) buff = io.BytesIO() img.save(buff, format=format) return buff.getvalue()
Converts numpy array to bytes in the specified image format. Args: np_img: numpy array H x W x C uint8 format: any format supported by Pillow, e.g. 'png' or 'jpeg' (note jpeg does not accept RGBA) Return: bytes
23,015
from __future__ import annotations from abc import ABC import sys import numpy as np import torch from glumpy import app, gloo, gl, ext import imgui from typing import Optional, Type, Callable, Dict, List, Tuple from kaolin.render.camera import Camera from wisp.framework import WispState, watch from wisp.renderer.core import RendererCore from wisp.renderer.core import cuda_register_gl_image, cuda_map_resource, cuda_2d_memcpy, cuda_unregister_resource from wisp.renderer.core.control import CameraControlMode, WispKey, WispMouseButton from wisp.renderer.core.control import FirstPersonCameraMode, TrackballCameraMode, TurntableCameraMode from wisp.renderer.core.api import add_pipeline_to_scene_graph from wisp.renderer.gizmos import Gizmo, WorldGrid, AxisPainter, PrimitivesPainter from wisp.renderer.gui import WidgetInteractiveVisualizerProperties, WidgetGPUStats, WidgetSceneGraph, WidgetImgui class WispApp(ABC): """ WispApp is a base app implementation which takes care of the entire lifecycle of the rendering loop: this is the infinite queue of events which includes: handling of IO and OS events, rendering frames and running backgrounds tasks, i.e. optimizations. The app is initiated by calling the following functions: - register_background_task(): Registers a task to run alternately with the render() function per frame. Background tasks can be, i.e, functions that run a single optimization step for some neural object. - run(): Initiates the rendering loop. This method blocks the calling thread until the window is closed. Future custom interactive apps can subclass this base app, and inherit methods to customize the behaviour of the app: - init_wisp_state(): A hook for initializing the fields of the shared state object. The interactive renderer configuration, scene graph and user custom fields can be initialized here. - update_render_state(): A hook for updating the fields of the shared state object at the beginning of each frame. - create_widgets(): Controls which gui components the app uses - create_gizmos(): Controls which transient canvas drawable components will be used (OpenGL based). - default_user_mode(): The default camera controller mode (first person, trackball, turntable). - register_event_handlers(): Registers which methods are invoked in response to app events / wisp state changes. The rendering loop alternates between the following calls: - on_idle, which invokes user background tasks, registered via register_background_task before calling run(). - on_draw, which invokes render() when it's time to draw a new frame Internally, the app uses the RendererCore object to manage the drawing of all objects in the scene graph. The app may request the RendererCore to switch into "interactive mode" to ensure the FPS remains interactive (this is done at the expense of rendering quality). Interactive mode is automatically initiated, i.e, during user interactions. The render() logic is composed by the following sub-functions: - update_render_state() - Updates the fields of the shared state object at the beginning of each frame. - render_gui() - Renders the imgui components over the canvas, fromm scratch (imgui uses immediate mode gui). - redraw() - A "heavier" function which forces the render-core to refresh its internal state. Newly created objects may get added to the scene graph, and obsolete objects may get removed. Vector-primitives data layers may regenerate here. - render_canvas() - Invokes the render-core to obtain a RenderBuffer of the rendered scene objects. The lion share of draw logic occurs within this call, in particular the drawing of neural objects. - _blit_to_gl_renderbuffer - Copies the RenderBuffer results to the screen buffer. - Gizmos are finally drawn directly to the screen framebuffer (as common OpenGL draw calls), these objects are considered transient in the sense that they don't belong to the scene graph. - Timer tick events may also get taken care of during the rendering loop (i.e: adjust velocity of user camera). Users should rarely override these functions, unless they're sure about what they're doing. Customizing the app behaviour should always be preferred via the initialization hooks. """ # Period of time between user interactions before resetting back to full resolution mode COOLDOWN_BETWEEN_RESOLUTION_CHANGES = 0.35 # In seconds def __init__(self, wisp_state, window_name="wisp app"): # Initialize app state instance self.wisp_state: WispState = wisp_state self.init_wisp_state(wisp_state) # Create main app window & initialize GL context # glumpy with a specialized glfw backend takes care of that (backend is imgui integration aware) window = self._create_window(self.width, self.height, window_name, gl_version=wisp_state.renderer.gl_version) self.register_io_mappings() # Initialize gui, assumes the window is managed by glumpy with glfw imgui.create_context() self._is_imgui_focused = False self._is_imgui_hovered = False self._is_reposition_imgui_menu = True self.canvas_dirty = False self.redraw_every_frame = False # Tell torch to initialize the CUDA context torch.cuda.init() # Initialize applicative renderer, which independently paints images for the main canvas render_core = RendererCore(self.wisp_state) self.window = window # App window with a GL context & oversees event callbacks self.render_core = render_core # Internal renderer, responsible for painting over canvas self.render_clock = app.clock.Clock() self.render_clock.tick() self.interactions_clock = app.clock.Clock() self.interactions_clock.tick() self._was_interacting_prev_frame = False # The initialization of these fields is deferred util "on_resize" is first prompted. # There we generate a simple billboard GL program (normally with a shared CUDA resource) # Canvas content will be blitted onto it self.canvas_program: Optional[gloo.Program] = None # GL program used to paint a single billboard self.cugl_rgb_handle = None # CUDA buffer, as a shared resource with OpenGL self.cugl_depth_handle = None try: # WSL does not support CUDA-OpenGL interoperability, fallback to device2host2device copy instead from platform import uname is_wsl = 'microsoft-standard' in uname().release self.blitdevice2device = not is_wsl except Exception: # By default rendering results copy directly from torch/cuda mem to OpenGL Texture self.blitdevice2device = True self.user_mode: CameraControlMode = None # Camera controller object (first person, trackball or turntable) self.widgets = self.create_widgets() # Create gui widgets for this app self.gizmos = self.create_gizmos() # Create canvas widgets for this app self.prim_painter = PrimitivesPainter() self.register_event_handlers() self.change_user_mode(self.default_user_mode()) self.redraw() # Refresh RendererCore def add_pipeline(self, name, pipeline, transform=None): """Register a neural fields pipeline into the scene graph. Args: name (str): The name of the pipeline. pipeline (wisp.models.Pipeline): The pipeline holding a tracer and nef. transform (wisp.core.ObjectTransform): The transform for the pipeline. """ add_pipeline_to_scene_graph(self.wisp_state, name, pipeline, transform=transform) def add_widget(self, widget, idx=None): """ Adds a widget to the list of widgets. By default, the widget is added to the end of the list of widgets, which means that it will be last on the display order. Args: widget (wisp.renderer.gui.imgui.WidgetImgui): The widget to add. idx (Optional[int]): If specified, will insert the widget at the specific display order. """ if idx is not None: self.widgets.insert(0, widget) else: self.widgets.append(widget) def add_gizmo(self, name, gizmo): """Adds a gizmo to the list of gizmos. Args: name (str): The name of the gizmo. gizmo (wisp.renderer.gizmos.Gizmo): The gizmo to add. """ self.gizmos[name] = gizmo def init_wisp_state(self, wisp_state: WispState) -> None: """ A hook for applications to initialize specific fields inside the wisp state object. This function is called at the very beginning of WispApp initialization, hence the initialized fields can be customized to affect the behaviour of the renderer. """ # Channels available to view over the canvas wisp_state.renderer.available_canvas_channels = ["rgb", "depth"] wisp_state.renderer.selected_canvas_channel = "rgb" def create_widgets(self) -> List[WidgetImgui]: """ Returns which widgets the gui will display, in order. Override to define which gui widgets are used by the wisp app. """ return [WidgetGPUStats(), WidgetInteractiveVisualizerProperties(), WidgetSceneGraph()] def create_gizmos(self) -> Dict[str, Gizmo]: """ Override to define which gizmos are painted on the canvas by the wisp app. Gizmos are transient rasterized objects rendered by OpenGL on top of the canvas. For example: world grid, axes painter. """ gizmos = dict() grid_size = 10.0 planes = self.wisp_state.renderer.reference_grids axes = set(''.join(planes)) for plane in planes: gizmos[f'world_grid_{plane}'] = WorldGrid(squares_per_axis=20, grid_size=grid_size, line_color=(128, 128, 128), line_size=1, plane=plane) gizmos[f'world_grid_fine_{plane}'] = WorldGrid(squares_per_axis=200, grid_size=10.0, line_color=(128, 128, 128), line_size=2, plane=plane) # Axes on top of the reference grid gizmos['grid_axis_painter'] = AxisPainter(axes_length=grid_size, line_width=1, axes=axes, is_bidirectional=False) return gizmos def default_user_mode(self) -> str: """ Override to determine the default camera control mode. Possible choices: 'First Person View', 'Turntable', 'Trackball' """ return "Turntable" def register_event_handlers(self) -> None: """ Override (and call super) to register additional event handlers """ watch(watched_obj=self.wisp_state.renderer, field="cam_controller", status="changed", handler=self.on_cam_controller_changed) watch(watched_obj=self.wisp_state.renderer, field="selected_camera", status="changed", handler=self.on_selected_camera_changed) watch(watched_obj=self.wisp_state.renderer, field="selected_canvas_channel", status="changed", handler=self.on_selected_canvas_channel_changed) watch(watched_obj=self.wisp_state.renderer, field="selected_camera_lens", status="changed", handler=self.on_selected_camera_lens_changed) watch(watched_obj=self.wisp_state.renderer, field="clear_color_value", status="changed", handler=self.on_clear_color_value_changed) def on_cam_controller_changed(self, value: Type[CameraControlMode]): # Stay synced with state change: generate new instance of mode type mode_type = value self.user_mode = mode_type(render_core=self.render_core, wisp_state=self.wisp_state) def on_selected_camera_changed(self, value: Camera): # Rebuild camera controller to free any cached info from previous camera cam_controller_cls = type(self.user_mode) self.user_mode = cam_controller_cls(render_core=self.render_core, wisp_state=self.wisp_state) # Adjust the width / height according to current state of the renderer window self.render_core.resize_canvas(height=self.height, width=self.width) def on_selected_camera_lens_changed(self, value: str): self.render_core.change_camera_projection_mode(value) def on_selected_canvas_channel_changed(self, value: str): self.canvas_dirty = True # Request canvas redraw def on_clear_color_value_changed(self, value: Tuple[float, float, float]): self.canvas_dirty = True # Request canvas redraw def run(self): """ Initiate events message queue, which triggers the rendering loop. This call will block the thread until the app window is closed. """ # glump.app.Window is using argparse # We remove sys.argv temporarily to avoid conflict with Wisp's argparse. argv is restored in on_init() self._argv = sys.argv sys.argv = [sys.argv[0]] try: app.run() # App clock should always run as frequently as possible (background tasks should not be limited) finally: if hasattr(self, '_argv'): # Should only be take place if an error occurred before on_init was invoked sys.argv = self._argv del self._argv def _create_window(self, width, height, window_name, gl_version): # glump.app.Window is using argparse # We remove sys.argv temporarily to avoid conflict with Wisp's argparse argv = sys.argv try: sys.argv = [argv[0]] # Currently assume glfw backend due to integration with imgui app.use(f"glfw_imgui ({gl_version})") win_config = app.configuration.Configuration() if self.wisp_state.renderer.antialiasing == 'msaa_4x': win_config.samples = 4 # glumpy implicitly sets the GL context as current window = app.Window(width=width, height=height, title=window_name, config=win_config) window.on_draw = self.on_draw window.on_resize = self.on_resize window.on_key_press = self.on_key_press window.on_key_release = self.on_key_release window.on_mouse_press = self.on_mouse_press window.on_mouse_drag = self.on_mouse_drag window.on_mouse_release = self.on_mouse_release window.on_mouse_scroll = self.on_mouse_scroll window.on_mouse_motion = self.on_mouse_motion finally: # Restore sys.argv for Wisp argparse sys.argv = argv if self.wisp_state.renderer.antialiasing == 'msaa_4x': gl.glEnable(gl.GL_MULTISAMPLE) return window def _create_gl_depth_billboard_program(texture: np.ndarray, depth_texture: np.ndarray): vertex = """ uniform float scale; attribute vec2 position; attribute vec2 texcoord; varying vec2 v_texcoord; void main() { v_texcoord = texcoord; gl_Position = vec4(scale*position, 0.0, 1.0); } """ fragment = """ uniform sampler2D tex; uniform sampler2D depth_tex; varying vec2 v_texcoord; void main() { gl_FragColor = texture2D(tex, v_texcoord); gl_FragDepth = texture2D(depth_tex, v_texcoord).r; } """ # TODO (operel): r component is a waste? # Compile GL program canvas = gloo.Program(vertex, fragment, count=4) # Upload fixed values to GPU canvas['position'] = [(-1, -1), (-1, +1), (+1, -1), (+1, +1)] canvas['texcoord'] = [(0, 0), (0, 1), (1, 0), (1, 1)] canvas['scale'] = 1.0 canvas['tex'] = texture canvas['depth_tex'] = depth_texture return canvas def _create_screen_texture(res_h, res_w, channel_depth, dtype=np.uint8): """ Create and return a Texture2D with gloo and a cuda handle. """ if issubclass(dtype, np.integer): tex = np.zeros((res_h, res_w, channel_depth), dtype).view(gloo.Texture2D) elif issubclass(dtype, np.floating): tex = np.zeros((res_h, res_w, channel_depth), dtype).view(gloo.TextureFloat2D) else: raise ValueError(f'_register_cugl_shared_texture invoked with unsupported texture dtype: {dtype}') # Force gloo to create GL object on GPU tex.activate() tex.deactivate() return tex def _register_cugl_shared_texture(self, tex): if self.blitdevice2device: # Create shared GL / CUDA resource handle = cuda_register_gl_image(image=int(tex.handle), target=tex.target) else: # No shared resource required, as we copy from cuda buffer -> cpu -> GL texture handle = None return handle def _reposition_gui_menu(self, menu_width, main_menu_height): window_height = self.window.height window_width = self.window.width imgui.set_next_window_size(width=menu_width, height=window_height-main_menu_height, condition=imgui.ALWAYS) imgui.set_next_window_position(x=window_width-menu_width, y=main_menu_height, condition=imgui.ALWAYS) self._is_reposition_imgui_menu = False def render_gui(self, state): """ Render the entire gui window per frame (imgui works in immediate mode). Internally, the Widgets take care of rendering the actual content. """ imgui.new_frame() if imgui.begin_main_menu_bar(): main_menu_height = imgui.get_window_height() if imgui.begin_menu("File", True): clicked_quit, selected_quit = imgui.menu_item( "Quit", 'Cmd+Q', False, True ) if clicked_quit: exit(1) imgui.end_menu() imgui.end_main_menu_bar() if self._is_reposition_imgui_menu: self._reposition_gui_menu(menu_width=350, main_menu_height=main_menu_height) imgui.begin("Scene Information", True) for widget in self.widgets: widget.paint(state) self._is_imgui_hovered = imgui.core.is_window_hovered(imgui.HOVERED_ANY_WINDOW | imgui.HOVERED_ALLOW_WHEN_BLOCKED_BY_POPUP | imgui.HOVERED_ALLOW_WHEN_BLOCKED_BY_ACTIVE_ITEM) self._is_imgui_hovered = self._is_imgui_hovered or \ imgui.core.is_any_item_hovered() or imgui.is_any_item_active() imgui.end() self._is_imgui_focused = imgui.is_window_focused(imgui.FOCUS_ROOT_WINDOW) imgui.end_frame() imgui.render() def render_canvas(self, render_core, time_delta, force_render): """ Invoke the render-core to render all neural fields and blend into a single Renderbuffer. The rgb and depth channels passed on to the app. """ # The render core returns a RenderBuffer renderbuffer = render_core.render(time_delta, force_render) buffer_attachment = renderbuffer.image().rgba buffer_attachment = buffer_attachment.flip([0]) # Flip y axis img = buffer_attachment.byte().contiguous() buffer_attachment_depth = renderbuffer.depth buffer_attachment_depth = buffer_attachment_depth.flip([0]) depth_img = buffer_attachment_depth.repeat(1,1,4).contiguous() return img, depth_img def _blit_to_gl_renderbuffer(self, img, depth_img, canvas_program, cugl_rgb_handle, cugl_depth_handle, height): if self.blitdevice2device: # Device to device copy: Copy CUDA buffer to GL Texture mem shared_tex = canvas_program['tex'] shared_tex_depth = canvas_program['depth_tex'] # copy from torch into buffer assert shared_tex.nbytes == img.numel() * img.element_size() assert shared_tex_depth.nbytes == depth_img.numel() * depth_img.element_size() # TODO: using a 4d tex cuda_2d_memcpy(resource_handle=cugl_rgb_handle, shared_tex=shared_tex, img=img, height=height) cuda_2d_memcpy(resource_handle=cugl_depth_handle, shared_tex=shared_tex_depth, img=depth_img, height=height) torch.cuda.synchronize() else: # Device to host to device copy: Move torch tensors to cpu and upload as texture data canvas_program['tex'] = img.cpu().numpy() canvas_program['depth_tex'] = depth_img.cpu().numpy() canvas_program.draw(gl.GL_TRIANGLE_STRIP) def update_renderer_state(self, wisp_state, dt): """ Populate the scene state object with the most recent information about the interactive renderer. The scene state, for example, may be used by the GUI widgets to display up to date information. This function is invoked in the beginning of the render() function, before the gui and the canvas are drawn. :param wisp_state The WispState object holding shared information between components about the wisp app. :param dt Amount of time elapsed since the last update. """ wisp_state.renderer.fps = app.clock.get_fps() wisp_state.renderer.dt = dt wisp_state.renderer.cam_controller = type(self.user_mode) def change_user_mode(self, user_mode: str): """ Changes the camera controller mode """ if user_mode == 'Trackball': self.wisp_state.renderer.cam_controller = TrackballCameraMode elif user_mode == 'First Person View': self.wisp_state.renderer.cam_controller = FirstPersonCameraMode elif user_mode == 'Turntable': self.wisp_state.renderer.cam_controller = TurntableCameraMode def redraw(self): """ Asks the render core to redraw the scene: - The scene graph will be refreshed (new objects added will create their renderers if needed) - Data layers will regenerate according to up-to-date state. render() may internally invoke redraw() when the canvas is tagged as "dirty". A render() call is required to display changes caused by redraw() on the canvas. """ # Let the renderer redraw the data layers if needed self.render_core.redraw() # Regenerate the GL primitives according to up-to-date data layers layers_to_draw = self.render_core.active_data_layers() self.prim_painter.redraw(layers_to_draw) def render(self): """ Renders a single frame. """ dt = self.render_clock.tick() # Tick render clock: dt is now the exact time elapsed since last render # Populate the scene state with the most recent information about the interactive renderer. # The scene state, for example, may be used by the GUI widgets to display up to date information. self.update_renderer_state(self.wisp_state, dt) # Clear color / depth buffers before rendering the next frame clear_color = (*self.wisp_state.renderer.clear_color_value, 1.0) # RGBA self.window.clear(color=clear_color) # imgui renders first self.render_gui(self.wisp_state) if self.redraw_every_frame: self.canvas_dirty = True # The app was asked to redraw the scene, inform the render core if self.canvas_dirty: self.redraw() # Invoke the timer tick event, and let the camera controller update the state of any interactions # of the user which involve the time elapsed (i.e: velocity, acceleration of movements). self.user_mode.handle_timer_tick(dt) # Toggle interactive mode on or off if needed to maintain interactive FPS rate if self.user_mode.is_interacting(): self.render_core.set_low_resolution() else: # Allow a fraction of a second before turning full resolution on. # User interactions sometimes behave like a rapid burst of short and quick interactions. if self._was_interacting_prev_frame: self.interactions_clock.tick() time_since_last_interaction = self.interactions_clock.time() - self.interactions_clock.last_ts if time_since_last_interaction > self.COOLDOWN_BETWEEN_RESOLUTION_CHANGES: self.render_core.set_full_resolution() self._was_interacting_prev_frame = self.user_mode.is_interacting() # render canvas: core proceeds by invoking internal renderers tracers # output is rendered on a Renderbuffer object, backed by torch tensors img, depth_img = self.render_canvas(self.render_core, dt, self.canvas_dirty) # glumpy code injected within the pyimgui render loop to blit the renderer core output to the actual canvas # The torch buffers are copied by with cuda, connected as shared resources as 2d GL textures self._blit_to_gl_renderbuffer(img, depth_img, self.canvas_program, self.cugl_rgb_handle, self.cugl_depth_handle, self.height) # Finally, render OpenGL gizmos on the canvas. # This may include the world grid, or vectorial lines / points belonging to data layers camera = self.render_core.camera for gizmo in self.gizmos.values(): gizmo.render(camera) self.prim_painter.render(camera) self.canvas_dirty = False def register_background_task(self, hook: Callable[[], None]) -> None: """ Register a new callable function to run in conjunction with the rendering loop. The app will alternate between on_idle calls, invoking the background task, and on_draw calls, invoking the rendering itself, both occurring on the same thread. """ if hook is not None: def _run_hook(dt: float): if not self.wisp_state.renderer.background_tasks_paused: hook() self.window.on_idle = _run_hook def on_draw(self, dt=None): """ glumpy's event to draw the next frame. Invokes the render() function if needed. """ # dt arg comes from the app clock, the renderer clock is maintained separately from the background tasks # Interactive mode on, or interaction have just started is_interacting = self.wisp_state.renderer.interactive_mode or self.user_mode.is_interacting() if is_interacting or self.is_time_to_render(): self.render() # Render objects uploaded to GPU def is_time_to_render(self): time_since_last_render = self.render_clock.time() - self.render_clock.last_ts target_fps = self.wisp_state.renderer.target_fps if target_fps is None or ((target_fps > 0) and time_since_last_render >= (1 / target_fps)): return True return False def on_init(self): """ Invoked when the app first runs """ # glump.app.Window is using argparse # During app.run(), we remove sys.argv temporarily to avoid conflict with Wisp's argparse. # Restore it here: this event is invoked by glumpy when app.run() initialization is complete. if hasattr(self, '_argv'): sys.argv = self._argv del self._argv def on_resize(self, width, height): """ Invoked when the window is first created, or resized. A resize causes internal textures and buffers to regenerate according the window size. """ self.width = width self.height = height # Handle deallocation of shared resources if self.cugl_rgb_handle is not None: cuda_unregister_resource(self.cugl_rgb_handle) self.cugl_rgb_handle = None if self.cugl_depth_handle is not None: cuda_unregister_resource(self.cugl_depth_handle) self.cugl_depth_handle = None tex = self._create_screen_texture(height, width, self.channel_depth, dtype=np.uint8) depth_tex = self._create_screen_texture(height, width, 4, dtype=np.float32) # TODO: Single channel self.cugl_rgb_handle = self._register_cugl_shared_texture(tex) self.cugl_depth_handle = self._register_cugl_shared_texture(depth_tex) if self.canvas_program is None: self.canvas_program = self._create_gl_depth_billboard_program(texture=tex, depth_texture=depth_tex) else: if self.canvas_program['tex'] is not None: self.canvas_program['tex'].delete() if self.canvas_program['depth_tex'] is not None: self.canvas_program['depth_tex'].delete() self.canvas_program['tex'] = tex self.canvas_program['depth_tex'] = depth_tex self.render_core.resize_canvas(height=height, width=width) self.window.activate() gl.glViewport(0, 0, width, height) self._is_reposition_imgui_menu = True # Signal menu it needs to shift after resize def is_canvas_event(self): """ Is canvas in focus or any of imgui's windows """ return not self._is_imgui_focused def on_mouse_press(self, x, y, button): if self.is_canvas_event(): self.user_mode.handle_mouse_press(x, y, button) def on_mouse_drag(self, x, y, dx, dy, button): if self.is_canvas_event(): self.user_mode.handle_mouse_drag(x, y, dx, dy, button) def on_mouse_release(self, x, y, button): if self.is_canvas_event(): self.user_mode.handle_mouse_release(x, y, button) def on_mouse_scroll(self, x, y, dx, dy): """ The mouse wheel was scrolled by (dx,dy). """ if self.is_canvas_event(): self.user_mode.handle_mouse_scroll(x, y, dx, dy) def on_mouse_motion(self, x, y, dx, dy): """ The mouse was moved with no buttons held down. """ if self.is_canvas_event(): self.user_mode.handle_mouse_motion(x, y, dx, dy) def width(self): """ Returns the canvas width """ return self.wisp_state.renderer.canvas_width def width(self, value: int): """ Sets the canvas width """ self.wisp_state.renderer.canvas_width = value def height(self): """ Returns the canvas height """ return self.wisp_state.renderer.canvas_height def height(self, value: int): """ Sets the canvas height """ self.wisp_state.renderer.canvas_height = value def channel_depth(self): """ Returns the number of channels the screenbuffer uses for the color attachment """ return 4 # Assume the framebuffer keeps RGBA def canvas_dirty(self): """ Returns if the canvas is dirty, that is, the app requires a redraw() to stay in sync with external changes """ return self.wisp_state.renderer.canvas_dirty def canvas_dirty(self, value: bool): """ Marks the canvas as dirty, implying the app requires a redraw() to stay in sync with external changes """ self.wisp_state.renderer.canvas_dirty = value def _update_imgui_keys(self, symbol): # Normally glfw shouldn't be explicitly imported as glumpy uses it as backend. # However, here we are forced to do that to take care of missing key mappings import glfw keys = [glfw.KEY_BACKSPACE, glfw.KEY_DELETE, glfw.KEY_ENTER, glfw.KEY_HOME, glfw.KEY_END, glfw.KEY_LEFT_SHIFT, glfw.KEY_RIGHT_SHIFT, glfw.KEY_RIGHT, glfw.KEY_LEFT, glfw.KEY_UP, glfw.KEY_DOWN, glfw.KEY_0, glfw.KEY_1, glfw.KEY_2, glfw.KEY_3, glfw.KEY_4, glfw.KEY_5, glfw.KEY_6, glfw.KEY_7, glfw.KEY_8, glfw.KEY_9, glfw.KEY_KP_0, glfw.KEY_KP_1, glfw.KEY_KP_2, glfw.KEY_KP_3, glfw.KEY_KP_4, glfw.KEY_KP_5, glfw.KEY_KP_6, glfw.KEY_KP_7, glfw.KEY_KP_8, glfw.KEY_KP_9, glfw.KEY_KP_ENTER, glfw.KEY_KP_ADD, glfw.KEY_KP_SUBTRACT, glfw.KEY_KP_DECIMAL ] mappings = { # Missing keys from glumpy glfw-imgui backend glfw.KEY_KP_0: glfw.KEY_0, glfw.KEY_KP_1: glfw.KEY_1, glfw.KEY_KP_2: glfw.KEY_2, glfw.KEY_KP_3: glfw.KEY_3, glfw.KEY_KP_4: glfw.KEY_4, glfw.KEY_KP_5: glfw.KEY_5, glfw.KEY_KP_6: glfw.KEY_6, glfw.KEY_KP_7: glfw.KEY_7, glfw.KEY_KP_8: glfw.KEY_8, glfw.KEY_KP_9: glfw.KEY_9, glfw.KEY_KP_ENTER: glfw.KEY_ENTER, glfw.KEY_KP_SUBTRACT: glfw.KEY_MINUS, glfw.KEY_KP_DECIMAL: glfw.KEY_PERIOD } updated_symbol = symbol for key in keys: is_key_on = glfw.get_key(self.window.native_window, key) imgui.get_io().keys_down[key] = is_key_on if symbol == -1 and is_key_on and key in mappings: updated_symbol = mappings[key] # TODO: Verify imgui keys have been properly mapped during initialization.. # imgui.get_io().key_map[imgui.KEY_BACKSPACE] = app.window.key.BACKSPACE # imgui.get_io().key_map[imgui.KEY_DELETE] = app.window.key.DELETE # imgui.get_io().key_map[imgui.KEY_ENTER] = glfw.KEY_ENTER # imgui.get_io().key_map[imgui.KEY_RIGHT_ARROW] = glfw.KEY_RIGHT # imgui.get_io().key_map[imgui.KEY_LEFT_ARROW] = glfw.KEY_LEFT # imgui.get_io().key_map[imgui.KEY_UP_ARROW] = glfw.KEY_UP # imgui.get_io().key_map[imgui.KEY_DOWN_ARROW] = glfw.KEY_DOWN return updated_symbol def on_key_press(self, symbol, modifiers): symbol = self._update_imgui_keys(symbol) if symbol > 0: imgui.get_io().add_input_character(symbol) if self.is_canvas_event(): self.user_mode.handle_key_press(symbol, modifiers) # TODO: Shouldn't be here cam_mode = None if symbol in (app.window.key.T, ord('T'), ord('t')): cam_mode = "Trackball" elif symbol in (app.window.key.F, ord('F'), ord('f')): cam_mode = "First Person View" elif symbol in (app.window.key.U, ord('U'), ord('u')): cam_mode = "Turntable" if cam_mode is not None: self.change_user_mode(cam_mode) def on_key_release(self, symbol, modifiers): symbol = self._update_imgui_keys(symbol) if self.is_canvas_event(): self.user_mode.handle_key_release(symbol, modifiers) def dump_framebuffer(self, path='./framebuffer'): # Dumps debug images of the GL screen framebuffer. # This framebuffer should reflect the exact content of the window. framebuffer = np.zeros((self.width, self.height * 3), dtype=np.uint8) gl.glReadPixels(0, 0, self.width, self.height, gl.GL_RGB, gl.GL_UNSIGNED_BYTE, framebuffer) framebuffer = np.flip(framebuffer, 0) ext.png.from_array(framebuffer, 'RGB').save(path + '_color.png') framebuffer = np.zeros((self.width, self.height), dtype=np.float32) gl.glReadPixels(0, 0, self.width, self.height, gl.GL_DEPTH_COMPONENT, gl.GL_FLOAT, framebuffer) framebuffer = np.flip(framebuffer, 0) ext.png.from_array(framebuffer, 'L').save(path + '_depth.png') def register_io_mappings(self): WispMouseButton.register_symbol(WispMouseButton.LEFT_BUTTON, app.window.mouse.LEFT) WispMouseButton.register_symbol(WispMouseButton.MIDDLE_BUTTON, app.window.mouse.MIDDLE) WispMouseButton.register_symbol(WispMouseButton.RIGHT_BUTTON, app.window.mouse.RIGHT) WispKey.register_symbol(WispKey.LEFT, app.window.key.LEFT) WispKey.register_symbol(WispKey.RIGHT, app.window.key.RIGHT) WispKey.register_symbol(WispKey.UP, app.window.key.UP) WispKey.register_symbol(WispKey.DOWN, app.window.key.DOWN) # TODO: Take care of remaining mappings, and verify the event handlers of glumpy were not overriden The provided code snippet includes necessary dependencies for implementing the `enable_amp` function. Write a Python function `def enable_amp(func)` to solve the following problem: An extension to @torch.cuda.amp.autocast which queries WispState to check if mixed precision should be enabled. Here is the function: def enable_amp(func): """ An extension to @torch.cuda.amp.autocast which queries WispState to check if mixed precision should be enabled. """ def _enable_amp(self: WispApp, *args, **kwargs): with torch.cuda.amp.autocast(enabled=self.wisp_state.renderer.enable_amp): return func(self, *args, **kwargs) return _enable_amp
An extension to @torch.cuda.amp.autocast which queries WispState to check if mixed precision should be enabled.
23,016
from __future__ import annotations from abc import ABC, abstractmethod from typing import Dict, Type, Any from wisp.framework import WispState from collections import deque from wisp.core.colors import colors_generator, white, black, dark_gray, gray _WIDGETS_REGISTRY: Dict[Type[Any], Type[WidgetImgui]] = dict() class WidgetImgui(ABC): def __init__(self): pass def paint(self, state: WispState, *args, **kwargs): raise NotImplementedError('imgui widgets must implement the paint method.') The provided code snippet includes necessary dependencies for implementing the `widget` function. Write a Python function `def widget(wisp_block: Type[Any])` to solve the following problem: A decorator that registers a gui widget to paint the contents of a given wisp block. By registering a widget, the gui system knows how to load this widget when it traverses the scene graph & properties and encounters the wisp_block type. Users adding new wisp blocks can directly register corresponding widgets using this decorator. Here is the function: def widget(wisp_block: Type[Any]): """ A decorator that registers a gui widget to paint the contents of a given wisp block. By registering a widget, the gui system knows how to load this widget when it traverses the scene graph & properties and encounters the wisp_block type. Users adding new wisp blocks can directly register corresponding widgets using this decorator. """ def _register_widget_fn(widget_class: Type[WidgetImgui]): _WIDGETS_REGISTRY[wisp_block] = widget_class return widget_class return _register_widget_fn
A decorator that registers a gui widget to paint the contents of a given wisp block. By registering a widget, the gui system knows how to load this widget when it traverses the scene graph & properties and encounters the wisp_block type. Users adding new wisp blocks can directly register corresponding widgets using this decorator.
23,017
from __future__ import annotations from abc import ABC, abstractmethod from typing import Dict, Type, Any from wisp.framework import WispState from collections import deque from wisp.core.colors import colors_generator, white, black, dark_gray, gray _WIDGETS_REGISTRY: Dict[Type[Any], Type[WidgetImgui]] = dict() class WidgetImgui(ABC): def __init__(self): pass def paint(self, state: WispState, *args, **kwargs): raise NotImplementedError('imgui widgets must implement the paint method.') def _lookup_widget(wisp_block: Any) -> Type[WidgetImgui]: """ Searches the wisp_block class hierarchy for any matches with some registered widget. wisp_block can be a """ # Create a queue of base classes and search if any of them has a registered widget. # Basically this is a BFS on the base classes, start with the concrete class type.. if not isinstance(wisp_block, type): # Is this a type or an instance? Fetch the type wisp_block = type(wisp_block) base_queue = deque([wisp_block]) widget_type = None while widget_type is None and base_queue: block_type = base_queue.popleft() widget_type = _WIDGETS_REGISTRY.get(block_type) if widget_type is None: bases = block_type.__bases__ if len(bases) > 0: base_queue.extend(bases) return widget_type The provided code snippet includes necessary dependencies for implementing the `get_widget` function. Write a Python function `def get_widget(wisp_block: Any) -> WidgetImgui` to solve the following problem: Return a widget which matches the given wisp block. A wisp block can be of any type / subtype which was registered with @widget. The lookup logic will first look for a widget registered under the type of wisp_block, and if it cannot find it, it will start looking up the hierarchy. Note that multiple-inheritance may result in undefined behavior in case of more than one match - in such cases users are encouraged to register some widget to the concrete type. Most wisp interfaces should already be registered to some general widget, so concrete types are not required to register a dedicated widget. However, they may register a specialized widget returned to display other forms of information. For example: Args: wisp_block: any object whose type / base types were registered with @widget. Returns: (WidgetImgui) imgui widget which matches the given wisp_block type Here is the function: def get_widget(wisp_block: Any) -> WidgetImgui: """ Return a widget which matches the given wisp block. A wisp block can be of any type / subtype which was registered with @widget. The lookup logic will first look for a widget registered under the type of wisp_block, and if it cannot find it, it will start looking up the hierarchy. Note that multiple-inheritance may result in undefined behavior in case of more than one match - in such cases users are encouraged to register some widget to the concrete type. Most wisp interfaces should already be registered to some general widget, so concrete types are not required to register a dedicated widget. However, they may register a specialized widget returned to display other forms of information. For example: Args: wisp_block: any object whose type / base types were registered with @widget. Returns: (WidgetImgui) imgui widget which matches the given wisp_block type """ block_type = type(wisp_block) widget_type = _WIDGETS_REGISTRY.get(block_type) if widget_type is None: widget_type = _lookup_widget(block_type) _WIDGETS_REGISTRY[block_type] = widget_type if widget_type is None: raise ValueError(f'Gui cannot find a widget for {widget_type}. Make sure to register some widget with @widget, ' f'or consider if {widget_type} should subclass WispModule.') widget_instance = widget_type() # Component widgets are assumed to take no args during construction return widget_instance
Return a widget which matches the given wisp block. A wisp block can be of any type / subtype which was registered with @widget. The lookup logic will first look for a widget registered under the type of wisp_block, and if it cannot find it, it will start looking up the hierarchy. Note that multiple-inheritance may result in undefined behavior in case of more than one match - in such cases users are encouraged to register some widget to the concrete type. Most wisp interfaces should already be registered to some general widget, so concrete types are not required to register a dedicated widget. However, they may register a specialized widget returned to display other forms of information. For example: Args: wisp_block: any object whose type / base types were registered with @widget. Returns: (WidgetImgui) imgui widget which matches the given wisp_block type
23,018
from __future__ import annotations import torch from kaolin.render.camera import Camera from kaolin.render.camera.intrinsics import CameraFOV from wisp.core import Rays def generate_default_grid(width, height, device=None): h_coords = torch.arange(height, device=device, dtype=torch.float) w_coords = torch.arange(width, device=device, dtype=torch.float) return torch.meshgrid(h_coords, w_coords) # return pixel_y, pixel_x def generate_centered_pixel_coords(img_width, img_height, res_x=None, res_y=None, device=None): pixel_y, pixel_x = generate_default_grid(res_x, res_y, device) scale_x = 1.0 if res_x is None else float(img_width) / res_x scale_y = 1.0 if res_y is None else float(img_height) / res_y pixel_x = pixel_x * scale_x + 0.5 # scale and add bias to pixel center pixel_y = pixel_y * scale_y + 0.5 # scale and add bias to pixel center return pixel_y, pixel_x
null
23,019
from __future__ import annotations import torch from kaolin.render.camera import Camera from kaolin.render.camera.intrinsics import CameraFOV from wisp.core import Rays def _to_ndc_coords(pixel_x, pixel_y, camera): pixel_x = 2 * (pixel_x / camera.width) - 1.0 pixel_y = 2 * (pixel_y / camera.height) - 1.0 return pixel_x, pixel_y The provided code snippet includes necessary dependencies for implementing the `generate_pinhole_rays` function. Write a Python function `def generate_pinhole_rays(camera: Camera, coords_grid: torch.Tensor)` to solve the following problem: Default ray generation function for pinhole cameras. This function assumes that the principal point (the pinhole location) is specified by a displacement (camera.x0, camera.y0) in pixel coordinates from the center of the image. The Kaolin camera class does not enforce a coordinate space for how the principal point is specified, so users will need to make sure that the correct principal point conventions are followed for the cameras passed into this function. Args: camera (kaolin.render.camera): The camera class. coords_grid (torch.FloatTensor): Grid of coordinates of shape [H, W, 2]. Returns: (wisp.core.Rays): The generated pinhole rays for the camera. Here is the function: def generate_pinhole_rays(camera: Camera, coords_grid: torch.Tensor): """Default ray generation function for pinhole cameras. This function assumes that the principal point (the pinhole location) is specified by a displacement (camera.x0, camera.y0) in pixel coordinates from the center of the image. The Kaolin camera class does not enforce a coordinate space for how the principal point is specified, so users will need to make sure that the correct principal point conventions are followed for the cameras passed into this function. Args: camera (kaolin.render.camera): The camera class. coords_grid (torch.FloatTensor): Grid of coordinates of shape [H, W, 2]. Returns: (wisp.core.Rays): The generated pinhole rays for the camera. """ if camera.device != coords_grid[0].device: raise Exception(f"Expected camera and coords_grid[0] to be on the same device, but found {camera.device} and {coords_grid[0].device}.") if camera.device != coords_grid[1].device: raise Exception(f"Expected camera and coords_grid[1] to be on the same device, but found {camera.device} and {coords_grid[1].device}.") # coords_grid should remain immutable (a new tensor is implicitly created here) pixel_y, pixel_x = coords_grid pixel_x = pixel_x.to(camera.device, camera.dtype) pixel_y = pixel_y.to(camera.device, camera.dtype) # Account for principal point (offsets from the center) pixel_x = pixel_x - camera.x0 pixel_y = pixel_y + camera.y0 # pixel values are now in range [-1, 1], both tensors are of shape res_y x res_x pixel_x, pixel_y = _to_ndc_coords(pixel_x, pixel_y, camera) ray_dir = torch.stack((pixel_x * camera.tan_half_fov(CameraFOV.HORIZONTAL), -pixel_y * camera.tan_half_fov(CameraFOV.VERTICAL), -torch.ones_like(pixel_x)), dim=-1) ray_dir = ray_dir.reshape(-1, 3) # Flatten grid rays to 1D array ray_orig = torch.zeros_like(ray_dir) # Transform from camera to world coordinates ray_orig, ray_dir = camera.extrinsics.inv_transform_rays(ray_orig, ray_dir) ray_dir /= torch.linalg.norm(ray_dir, dim=-1, keepdim=True) ray_orig, ray_dir = ray_orig[0], ray_dir[0] # Assume a single camera return Rays(origins=ray_orig, dirs=ray_dir, dist_min=camera.near, dist_max=camera.far)
Default ray generation function for pinhole cameras. This function assumes that the principal point (the pinhole location) is specified by a displacement (camera.x0, camera.y0) in pixel coordinates from the center of the image. The Kaolin camera class does not enforce a coordinate space for how the principal point is specified, so users will need to make sure that the correct principal point conventions are followed for the cameras passed into this function. Args: camera (kaolin.render.camera): The camera class. coords_grid (torch.FloatTensor): Grid of coordinates of shape [H, W, 2]. Returns: (wisp.core.Rays): The generated pinhole rays for the camera.
23,020
from __future__ import annotations import torch from kaolin.render.camera import Camera from kaolin.render.camera.intrinsics import CameraFOV from wisp.core import Rays def _to_ndc_coords(pixel_x, pixel_y, camera): pixel_x = 2 * (pixel_x / camera.width) - 1.0 pixel_y = 2 * (pixel_y / camera.height) - 1.0 return pixel_x, pixel_y def generate_ortho_rays(camera: Camera, coords_grid: torch.Tensor): # coords_grid should remain immutable (a new tensor is implicitly created here) pixel_y, pixel_x = coords_grid pixel_y = pixel_y.to(camera.device, camera.dtype) pixel_x = pixel_x.to(camera.device, camera.dtype) # pixel values are now in range [-1, 1], both tensors are of shape res_y x res_x pixel_x, pixel_y = _to_ndc_coords(pixel_x, pixel_y, camera) # Rescale according to distance from camera aspect_ratio = camera.width / camera.height pixel_x *= camera.fov_distance * aspect_ratio pixel_y *= camera.fov_distance zeros = torch.zeros_like(pixel_x) ray_dir = torch.stack((zeros, zeros, -torch.ones_like(pixel_x)), dim=-1) # Ortho rays are parallel ray_orig = torch.stack((pixel_x, -pixel_y, zeros), dim=-1) ray_dir = ray_dir.reshape(-1, 3) # Flatten grid rays to 1D array ray_orig = ray_orig.reshape(-1, 3) # Flatten grid rays to 1D array # Transform from camera to world coordinates ray_orig, ray_dir = camera.extrinsics.inv_transform_rays(ray_orig, ray_dir) ray_dir /= torch.linalg.norm(ray_dir, dim=-1, keepdim=True) ray_orig, ray_dir = ray_orig[0], ray_dir[0] # Assume a single camera return Rays(origins=ray_orig, dirs=ray_dir, dist_min=camera.near, dist_max=camera.far)
null
23,021
import torch The provided code snippet includes necessary dependencies for implementing the `autodiff_gradient` function. Write a Python function `def autodiff_gradient(x, f)` to solve the following problem: Compute gradient using the PyTorch autodiff. Args: x (torch.FloatTensor): Coordinate tensor f (nn.Module): The function to perform autodiff on. Here is the function: def autodiff_gradient(x, f): """Compute gradient using the PyTorch autodiff. Args: x (torch.FloatTensor): Coordinate tensor f (nn.Module): The function to perform autodiff on. """ with torch.enable_grad(): x = x.requires_grad_(True) y = f(x) grad = torch.autograd.grad(y, x, grad_outputs=torch.ones_like(y), create_graph=True)[0] return grad
Compute gradient using the PyTorch autodiff. Args: x (torch.FloatTensor): Coordinate tensor f (nn.Module): The function to perform autodiff on.
23,022
import torch The provided code snippet includes necessary dependencies for implementing the `finitediff_gradient` function. Write a Python function `def finitediff_gradient(x, f, eps=0.005)` to solve the following problem: Compute 3D gradient using finite difference. Args: x (torch.FloatTensor): Coordinate tensor of shape [..., 3] f (nn.Module): The function to perform autodiff on. Here is the function: def finitediff_gradient(x, f, eps=0.005): """Compute 3D gradient using finite difference. Args: x (torch.FloatTensor): Coordinate tensor of shape [..., 3] f (nn.Module): The function to perform autodiff on. """ eps_x = torch.tensor([eps, 0.0, 0.0], device=x.device) eps_y = torch.tensor([0.0, eps, 0.0], device=x.device) eps_z = torch.tensor([0.0, 0.0, eps], device=x.device) grad = torch.cat([f(x + eps_x) - f(x - eps_x), f(x + eps_y) - f(x - eps_y), f(x + eps_z) - f(x - eps_z)], dim=-1) grad = grad / (eps * 2.0) return grad
Compute 3D gradient using finite difference. Args: x (torch.FloatTensor): Coordinate tensor of shape [..., 3] f (nn.Module): The function to perform autodiff on.
23,023
import torch The provided code snippet includes necessary dependencies for implementing the `tetrahedron_gradient` function. Write a Python function `def tetrahedron_gradient(x, f, eps=0.005)` to solve the following problem: Compute 3D gradient using finite difference (using tetrahedron method). Args: x (torch.FloatTensor): Coordinate tensor of shape [..., 3] f (nn.Module): The function to perform autodiff on. Here is the function: def tetrahedron_gradient(x, f, eps=0.005): """Compute 3D gradient using finite difference (using tetrahedron method). Args: x (torch.FloatTensor): Coordinate tensor of shape [..., 3] f (nn.Module): The function to perform autodiff on. """ h = eps k0 = torch.tensor([1.0, -1.0, -1.0], device=x.device, requires_grad=False) k1 = torch.tensor([-1.0, -1.0, 1.0], device=x.device, requires_grad=False) k2 = torch.tensor([-1.0, 1.0, -1.0], device=x.device, requires_grad=False) k3 = torch.tensor([1.0, 1.0, 1.0], device=x.device, requires_grad=False) h0 = torch.tensor([h, -h, -h], device=x.device, requires_grad=False) h1 = torch.tensor([-h, -h, h], device=x.device, requires_grad=False) h2 = torch.tensor([-h, h, -h], device=x.device, requires_grad=False) h3 = torch.tensor([h, h, h], device=x.device, requires_grad=False) h0 = x + h0 h1 = x + h1 h2 = x + h2 h3 = x + h3 h0 = h0.detach() h1 = h1.detach() h2 = h2.detach() h3 = h3.detach() h0 = k0 * f(h0) h1 = k1 * f(h1) h2 = k2 * f(h2) h3 = k3 * f(h3) grad = (h0 + h1 + h2 + h3) / (h * 4.0) return grad
Compute 3D gradient using finite difference (using tetrahedron method). Args: x (torch.FloatTensor): Coordinate tensor of shape [..., 3] f (nn.Module): The function to perform autodiff on.
23,024
import torch from kaolin import _C import wisp._C as wisp_C import kaolin.ops.spc as spc_ops PRIMES = [1, 2654435761, 805459861] The provided code snippet includes necessary dependencies for implementing the `hashgrid_naive` function. Write a Python function `def hashgrid_naive(coords, resolutions, codebook_bitwidth, lod_idx, codebook, codebook_lod_sizes, codebook_lod_first_idx)` to solve the following problem: A naive PyTorch implementation of the hashgrid. This code exists here mostly as a reference: Do NOT expect a 1-to-1 numerical correspondence to the CUDA accelerated version. This code is comparatively very slow. :) Args: coords (torch.FloatTensor): 3D coordinates of shape [batch, 3] resolutions (torch.LongTensor): the resolution of the grid per level of shape [num_lods] codebook_bitwidth (int): The bitwidth of the codebook. The codebook will have 2^bw entries. lod_idx (int): The LOD to aggregate to. codebook (torch.FloatTensor): A tensor containing the stacked codebooks, each of shape [codebook_size_lod_idx, feature_dim]. codebook_lod_sizes (torch.IntTensor): A tensor containig the codebook size at each level of detail. codebook_lod_first_idx (torch.IntTensor): A tensor containing the first index of each codebook in the stacked codebook tensor. Returns: (torch.FloatTensor): Features of shape [batch*num_samples, feature_dim] Here is the function: def hashgrid_naive(coords, resolutions, codebook_bitwidth, lod_idx, codebook, codebook_lod_sizes, codebook_lod_first_idx): """ A naive PyTorch implementation of the hashgrid. This code exists here mostly as a reference: Do NOT expect a 1-to-1 numerical correspondence to the CUDA accelerated version. This code is comparatively very slow. :) Args: coords (torch.FloatTensor): 3D coordinates of shape [batch, 3] resolutions (torch.LongTensor): the resolution of the grid per level of shape [num_lods] codebook_bitwidth (int): The bitwidth of the codebook. The codebook will have 2^bw entries. lod_idx (int): The LOD to aggregate to. codebook (torch.FloatTensor): A tensor containing the stacked codebooks, each of shape [codebook_size_lod_idx, feature_dim]. codebook_lod_sizes (torch.IntTensor): A tensor containig the codebook size at each level of detail. codebook_lod_first_idx (torch.IntTensor): A tensor containing the first index of each codebook in the stacked codebook tensor. Returns: (torch.FloatTensor): Features of shape [batch*num_samples, feature_dim] """ codebook_size = 2**codebook_bitwidth feats = [] for i, res in enumerate(resolutions[:lod_idx+1]): # This assumes that the input coordinates are in the range [0, 1]. tf_coords = torch.clip(((coords + 1.0) / 2.0) * res, 0, res-1-1e-5).reshape(-1, 3) cc000 = torch.floor(tf_coords).short() cc = spc_ops.points_to_corners(cc000).long() num_pts = res**3 if num_pts > codebook_size: cidx = ( (cc[...,0] * PRIMES[0]) ^ (cc[...,1] * PRIMES[1]) ^ (cc[...,2] * PRIMES[2]) ) % codebook_size else: cidx = cc[...,0] + cc[...,1] * res + cc[...,2] * res * res # cidx: B, 8 fs = codebook[codebook_lod_first_idx[i] : codebook_lod_first_idx[i] + codebook_lod_sizes[i]][cidx.reshape(-1)] # B*8, F fs = fs.reshape(-1, 8, fs.shape[-1]) # B, 8, F coeffs = torch.zeros(coords.size(0), 8, device=coords.device, dtype=coords.dtype) # B, 8 x = tf_coords - cc000 _x = 1.0 - x # Trilinear interpolation coeffs[...,0] = _x[...,0] * _x[...,1] * _x[...,2] coeffs[...,1] = _x[...,0] * _x[...,1] * x[...,2] coeffs[...,2] = _x[...,0] * x[...,1] * _x[...,2] coeffs[...,3] = _x[...,0] * x[...,1] * x[...,2] coeffs[...,4] = x[...,0] * _x[...,1] * _x[...,2] coeffs[...,5] = x[...,0] * _x[...,1] * x[...,2] coeffs[...,6] = x[...,0] * x[...,1] * _x[...,2] coeffs[...,7] = x[...,0] * x[...,1] * x[...,2] coeffs = coeffs.reshape(-1, 8, 1) # B, 8, 1 fs_coeffs = (fs * coeffs).sum(1) # B, F feats.append(fs_coeffs) # TODO(ttakikawa): This probably does not return according to the num_samples interface return torch.cat(feats, -1) # B, F*L
A naive PyTorch implementation of the hashgrid. This code exists here mostly as a reference: Do NOT expect a 1-to-1 numerical correspondence to the CUDA accelerated version. This code is comparatively very slow. :) Args: coords (torch.FloatTensor): 3D coordinates of shape [batch, 3] resolutions (torch.LongTensor): the resolution of the grid per level of shape [num_lods] codebook_bitwidth (int): The bitwidth of the codebook. The codebook will have 2^bw entries. lod_idx (int): The LOD to aggregate to. codebook (torch.FloatTensor): A tensor containing the stacked codebooks, each of shape [codebook_size_lod_idx, feature_dim]. codebook_lod_sizes (torch.IntTensor): A tensor containig the codebook size at each level of detail. codebook_lod_first_idx (torch.IntTensor): A tensor containing the first index of each codebook in the stacked codebook tensor. Returns: (torch.FloatTensor): Features of shape [batch*num_samples, feature_dim]
23,025
import torch from kaolin import _C import wisp._C as wisp_C import kaolin.ops.spc as spc_ops class HashGridInterpolate(torch.autograd.Function): # TODO(ttakikawa): This class should also support the 2D case... which also means I have to write another kernel! def forward(ctx, coords, resolutions, codebook_bitwidth, lod_idx, codebook, codebook_first_idx): if codebook[0].shape[-1] % 2 == 1: raise Exception("The codebook feature dimension needs to be a multiple of 2.") assert(coords.shape[-1] in [2, 3]) if torch.is_autocast_enabled(): codebook = codebook.half() # TODO(ttakikawa): Make the kernel use the LOD feats_out = wisp_C.ops.hashgrid_interpolate_cuda(coords.contiguous(), codebook, codebook_first_idx, resolutions, codebook_bitwidth).contiguous() ctx.save_for_backward(coords, codebook, codebook_first_idx) ctx.resolutions = resolutions ctx.num_lods = len(resolutions) ctx.codebook_size = 2**codebook_bitwidth ctx.codebook_bitwidth = codebook_bitwidth ctx.feature_dim = codebook.shape[-1] return feats_out def backward(ctx, grad_output): coords = ctx.saved_tensors[0] codebook = ctx.saved_tensors[1] codebook_first_idx = ctx.saved_tensors[2] resolutions = ctx.resolutions feature_dim = ctx.feature_dim codebook_bitwidth = ctx.codebook_bitwidth coords_requires_grad = ctx.needs_input_grad[0] grad_coords, grad_codebook = wisp_C.ops.hashgrid_interpolate_backward_cuda( coords.float().contiguous(), grad_output.contiguous(), codebook, codebook_first_idx, resolutions, codebook_bitwidth, feature_dim, coords_requires_grad) if coords_requires_grad: return (grad_coords, None, None, None, grad_codebook, None, None) else: return (None, None, None, None, grad_codebook, None, None) The provided code snippet includes necessary dependencies for implementing the `hashgrid` function. Write a Python function `def hashgrid(coords, codebook_bitwidth, lod_idx, codebook)` to solve the following problem: A hash-grid query + interpolation function, accelerated with CUDA. Args: coords (torch.FloatTensor): 3D coordinates of shape [batch, 3] codebook_bitwidth (int): The bitwidth of the codebook. The codebook will have 2^bw entries. lod_idx (int): The LOD to aggregate to. codebook (wisp.models.grids.utils.MultiTable): A class that holds multiresolution tables. Returns: (torch.FloatTensor): Features of shape [batch, feature_dim] Here is the function: def hashgrid(coords, codebook_bitwidth, lod_idx, codebook): """A hash-grid query + interpolation function, accelerated with CUDA. Args: coords (torch.FloatTensor): 3D coordinates of shape [batch, 3] codebook_bitwidth (int): The bitwidth of the codebook. The codebook will have 2^bw entries. lod_idx (int): The LOD to aggregate to. codebook (wisp.models.grids.utils.MultiTable): A class that holds multiresolution tables. Returns: (torch.FloatTensor): Features of shape [batch, feature_dim] """ batch, dim = coords.shape feats = HashGridInterpolate.apply(coords.contiguous(), codebook.resolutions, codebook_bitwidth, lod_idx, codebook.feats, codebook.begin_idxes) feature_dim = codebook.feats.shape[1] * len(codebook.resolutions) return feats.reshape(batch, feature_dim)
A hash-grid query + interpolation function, accelerated with CUDA. Args: coords (torch.FloatTensor): 3D coordinates of shape [batch, 3] codebook_bitwidth (int): The bitwidth of the codebook. The codebook will have 2^bw entries. lod_idx (int): The LOD to aggregate to. codebook (wisp.models.grids.utils.MultiTable): A class that holds multiresolution tables. Returns: (torch.FloatTensor): Features of shape [batch, feature_dim]