def render(self, mode='rgb_array', *args, **kwargs):
"""
@brief Parallel rendering of the current state of each environment.
@param[in] mode Display mode. For now, only 'rgb_array' is supported.
@param[in] args Extra arguments passed to `render` method of each Gym environment
@param[in] kwargs Extra keyword arguments passed to `render` method of each Gym environment
@return None or iterator of RGB images if mode == 'rgb_array'.
"""
# gather images from subprocesses. `mode` will be taken into account later
for pipe in self.remotes:
# gather images from subprocesses
# `mode` will be taken into account later
pipe.send(('render', (args, {'mode': 'rgb_array', **kwargs})))
imgs = [pipe.recv() for pipe in self.remotes]
# Create a big image by tiling images from subprocesses
bigimg = tile_images(imgs)
if mode == 'rgb_array':
return bigimg
else:
raise NotImplementedError
def attention_render(model_name, env_name, num_cpu, log_dir):
if not os.path.exists(log_dir):
raise ('log_dir not Exists')
env_id = env_name + 'NoFrameskip-v4'
env = SubprocVecEnv([make_env(env_id, i, log_dir) for i in range(num_cpu)])
# env = Monitor(env, log_dir, allow_early_resets=True)
if model_name == 'A2C_Attention':
model = A2C(AttentionPolicy,
env,
verbose=1,
tensorboard_log=log_dir + 'tensorboard/')
elif model_name == 'A2C_Attention2':
model = A2C(Attention2Policy,
env,
verbose=1,
tensorboard_log=log_dir + 'tensorboard/')
elif model_name == 'A2C':
model = A2C(LstmPolicy,
env,
verbose=1,
tensorboard_log=log_dir + 'tensorboard/')
else:
model = None
model = model.load(log_dir + model_name + '_' + env_name, env=env)
obs = env.reset()
# print(env.observation_space)
# cv2.imshow('test', RGB2BGR(obs[0]))
# cv2.waitKey(0)
while True:
action, _states = model.predict(obs)
obs, rewards, done, info = env.step(action)
attentions = model.get_attention(obs, _states, done)[0]
attentions_img = []
# print('attention', np.array(attention).shape)
for i, attention in enumerate(attentions):
attention = np.array(attention)
attention = np.reshape(attention, [
env.observation_space.shape[0] // 10,
env.observation_space.shape[1] // 10, 1
])
attention = np.repeat(attention, [10] * attention.shape[0], axis=0)
attention = np.repeat(attention, [10] * attention.shape[1], axis=1)
attention = attention * 255
attentions_img.append(attention)
# print(np.sum(attention))
attentions = tile_images(attentions_img)
cv2.imshow('attention', attentions)
cv2.waitKey(1)
# break
env.render()
return model
def render(self, mode='human'):
for pipe in self.remotes:
pipe.send(('render', None))
imgs = [pipe.recv() for pipe in self.remotes]
bigimg = tile_images(imgs)
if mode == 'human':
import cv2
cv2.imshow('vecenv', bigimg[:, :, ::-1])
cv2.waitKey(1)
elif mode == 'rgb_array':
return bigimg
else:
raise NotImplementedError
def render(self, mode='human', *args, **kwargs):
# gather images from subprocesses
# `mode` will be taken into account later
imgs = self.get_images(*args, **kwargs)
# Create a big image by tiling images from subprocesses
bigimg = tile_images(imgs)
if mode == 'human':
import cv2
cv2.imshow('vecenv', bigimg[:, :, ::-1])
cv2.waitKey(1)
elif mode == 'rgb_array':
return bigimg
else:
raise NotImplementedError
def render(self, mode='human', *args, **kwargs):
for pipe in self.remotes:
# gather images from subprocesses
# `mode` will be taken into account later
pipe.send(('render', (args, {'mode': 'rgb_array', **kwargs})))
imgs = [pipe.recv() for pipe in self.remotes]
# Create a big image by tiling images from subprocesses
bigimg = tile_images(imgs)
if mode == 'human':
import cv2
cv2.imshow('vecenv', bigimg[:, :, ::-1])
cv2.waitKey(1)
elif mode == 'rgb_array':
return bigimg
else:
raise NotImplementedError
def render(self, indices=None, *args, **kwargs):
for pipe in self._get_target_remotes(indices):
# gather images from subprocesses
# `mode` will be taken into account later
pipe.send(('render', (args, {**kwargs})))
figs = [pipe.recv() for pipe in self._get_target_remotes(indices)]
if isinstance(indices, int):
figs = figs[0]
return figs
# Create a big image by tiling images from subprocesses
bigimg = tile_images(imgs)
if mode == 'human':
import cv2
cv2.imshow('vecenv', bigimg[:, :, ::-1])
cv2.waitKey(1)
elif mode == 'rgb_array':
return bigimg
else:
raise NotImplementedError
def _record(self, obs):
self.last += 1
if self.last - self.reduction < 0:
return
else:
self.last = 0
if self.record_obs:
if self.unvec:
for i, recorder in enumerate(self.recorders):
recorder.record(obs[i])
else:
self.recorders[0].record(tile_images(obs))
else:
if self.unvec:
images = self.venv.env_method("render", mode="rgb_array")
for i, recorder in enumerate(self.recorders):
recorder.record(images[i])
else:
self.recorders[0].record(self.venv.render(mode="rgb_array"))
def render(self, mode: str, *args, **kwargs):
"""
Gym environment rendering
:param mode: the rendering type
"""
try:
imgs = self.get_images(*args, **kwargs)
except NotImplementedError:
logger.warn('Render not defined for {}'.format(self))
return
# Create a big image by tiling images from subprocesses
bigimg = tile_images(imgs)
if mode == 'human':
import cv2 # pytype:disable=import-error
cv2.imshow('vecenv', bigimg[:, :, ::-1])
cv2.waitKey(1)
elif mode == 'rgb_array':
return bigimg
else:
raise NotImplementedError
