def save_hko_gif(im_dat, save_path):
"""Save the HKO images to gif
Parameters
----------
im_dat : np.ndarray
Shape: (seqlen, H, W)
save_path : str
Returns
-------
"""
assert im_dat.ndim == 3
save_gif(im_dat, fname=save_path)
return
def analysis(args):
cfg.MODEL.TRAJRNN.SAVE_MID_RESULTS = True
assert cfg.MODEL.FRAME_STACK == 1 and cfg.MODEL.FRAME_SKIP == 1
base_dir = args.save_dir
logging_config(folder=base_dir, name="testing")
save_movingmnist_cfg(base_dir)
mnist_iter = MovingMNISTAdvancedIterator(
distractor_num=cfg.MOVINGMNIST.DISTRACTOR_NUM,
initial_velocity_range=(cfg.MOVINGMNIST.VELOCITY_LOWER,
cfg.MOVINGMNIST.VELOCITY_UPPER),
rotation_angle_range=(cfg.MOVINGMNIST.ROTATION_LOWER,
cfg.MOVINGMNIST.ROTATION_UPPER),
scale_variation_range=(cfg.MOVINGMNIST.SCALE_VARIATION_LOWER,
cfg.MOVINGMNIST.SCALE_VARIATION_UPPER),
illumination_factor_range=(cfg.MOVINGMNIST.ILLUMINATION_LOWER,
cfg.MOVINGMNIST.ILLUMINATION_UPPER))
mnist_rnn = MovingMNISTFactory(batch_size=1,
in_seq_len=cfg.MODEL.IN_LEN,
out_seq_len=cfg.MODEL.OUT_LEN)
encoder_net, forecaster_net, loss_net = \
encoder_forecaster_build_networks(
factory=mnist_rnn,
context=args.ctx)
encoder_net.summary()
forecaster_net.summary()
loss_net.summary()
states = EncoderForecasterStates(factory=mnist_rnn, ctx=args.ctx[0])
states.reset_all()
# Begin to load the model if load_dir is not empty
assert len(cfg.MODEL.LOAD_DIR) > 0
load_encoder_forecaster_params(load_dir=cfg.MODEL.LOAD_DIR,
load_iter=cfg.MODEL.LOAD_ITER,
encoder_net=encoder_net,
forecaster_net=forecaster_net)
for iter_id in range(1):
frame_dat, _ = mnist_iter.sample(batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
seqlen=cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN)
data_nd = mx.nd.array(frame_dat[0:cfg.MOVINGMNIST.IN_LEN, ...],
ctx=args.ctx[0]) / 255.0
target_nd = mx.nd.array(frame_dat[cfg.MOVINGMNIST.IN_LEN:(
cfg.MOVINGMNIST.IN_LEN + cfg.MOVINGMNIST.OUT_LEN), ...],
ctx=args.ctx[0]) / 255.0
pred_nd = mnist_get_prediction(data_nd=data_nd,
states=states,
encoder_net=encoder_net,
forecaster_net=forecaster_net)
save_gif(pred_nd.asnumpy()[:, 0, 0, :, :],
os.path.join(base_dir, "pred.gif"))
save_gif(data_nd.asnumpy()[:, 0, 0, :, :],
os.path.join(base_dir, "in.gif"))
save_gif(target_nd.asnumpy()[:, 0, 0, :, :],
os.path.join(base_dir, "gt.gif"))
def train(args):
base_dir = get_base_dir(args)
### Get modules
generator_net, loss_net = construct_modules(args)
### Prepare data
mnist_iter = MovingMNISTAdvancedIterator(
distractor_num=cfg.MOVINGMNIST.DISTRACTOR_NUM,
initial_velocity_range=(cfg.MOVINGMNIST.VELOCITY_LOWER,
cfg.MOVINGMNIST.VELOCITY_UPPER),
rotation_angle_range=(cfg.MOVINGMNIST.ROTATION_LOWER,
cfg.MOVINGMNIST.ROTATION_UPPER),
scale_variation_range=(cfg.MOVINGMNIST.SCALE_VARIATION_LOWER,
cfg.MOVINGMNIST.SCALE_VARIATION_UPPER),
illumination_factor_range=(cfg.MOVINGMNIST.ILLUMINATION_LOWER,
cfg.MOVINGMNIST.ILLUMINATION_UPPER))
for i in range(cfg.MODEL.TRAIN.MAX_ITER):
seq, flow = mnist_iter.sample(batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
seqlen=cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN)
in_seq = seq[:cfg.MOVINGMNIST.IN_LEN, ...]
gt_seq = seq[cfg.MOVINGMNIST.IN_LEN:(cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN), ...]
# Transform data to NCDHW shape needed for 3D Convolution encoder and normalize
context_nd = mx.nd.array(in_seq) / 255.0
gt_nd = mx.nd.array(gt_seq) / 255.0
context_nd = mx.nd.transpose(context_nd, axes=(1, 2, 0, 3, 4))
gt_nd = mx.nd.transpose(gt_nd, axes=(1, 2, 0, 3, 4))
# Train a step
pred_nd, avg_l2, avg_real_mse, generator_grad_norm =\
train_step(generator_net, loss_net, context_nd, gt_nd)
# Logging
logging.info(
("Iter:{}, L2 Loss:{}, MSE Error:{}, Generator Grad Norm:{}"
).format(i, avg_l2, avg_real_mse, generator_grad_norm))
logging.info("Iter:%d" % i)
if (i + 1) % 100 == 0:
save_gif(context_nd.asnumpy()[0, 0, :, :, :],
os.path.join(base_dir, "input.gif"))
save_gif(gt_nd.asnumpy()[0, 0, :, :, :],
os.path.join(base_dir, "gt.gif"))
save_gif(pred_nd.asnumpy()[0, 0, :, :, :],
os.path.join(base_dir, "pred.gif"))
if cfg.MODEL.SAVE_ITER > 0 and (i + 1) % cfg.MODEL.SAVE_ITER == 0:
generator_net.save_checkpoint(prefix=os.path.join(
base_dir, "generator"),
epoch=i)
is_train=True)
outputs = net.get_outputs()
net.backward()
norm_val = get_global_norm_val(net)
# norm_clipping(params_grad=[grad[0] for grad in net._exec_group.grad_arrays],
# threshold=100, batch_size=batch_size)
logging.info(
"Iter:%d, Error:%f, Norm:%f" %
(i, outputs[0].asnumpy().sum() / batch_size / 64 / 64, norm_val))
for k, v, grad_v in zip(net._param_names, net._exec_group.param_arrays,
net._exec_group.grad_arrays):
if "bn" not in k:
print k, v[0].shape, nd.norm(v[0]).asnumpy(), nd.norm(
grad_v[0] / batch_size).asnumpy()
net.update()
if (i + 1) % 100 == 0:
test_net.forward(data_batch=mx.io.DataBatch(
data=[mx.nd.array(in_seq) / 255.0], label=None),
is_train=False)
test_prediction = test_net.get_outputs()[0].asnumpy()
logging.info(
"Iter:%d, Test Error:%f" %
(i, -cross_entropy_npy(gt_seq / 255.0, test_prediction).sum() /
out_seq_len / batch_size))
save_gif(test_prediction[:, 0, 0, :, :], "test.gif")
if (i + 1) % 2000 == 0:
net.save_checkpoint(prefix=os.path.join(
base_dir, "%s_%s" % (conv_rnn_typ, transform_typ)),
epoch=i)
batch_size = 1
if args.mode == 'test':
seqlen = 100
elif args.mode == 'save':
if args.path:
fname = args.path
else:
fname = "params.npz"
print("Generating {} sequences of length {}. Saving to {}.".format(
args.sequences, args.length, fname))
seqlen = args.length
mnist_generator.save(seqlen=seqlen,
num_samples=args.sequences,
file=fname)
elif args.mode == 'load':
if args.path:
fname = args.path
else:
fname = "params.npz"
num_sequences, seqlen = mnist_generator.load(file=fname)
print("Loaded {} sequences of length {}. Saving to {}.".format(
num_sequences, seqlen, fname))
seq, _ = mnist_generator.sample(batch_size=batch_size, seqlen=seqlen)
print(seq.sum())
save_gif(seq[:, 0, 0, :, :].astype(np.float32) / 255.0, "test.gif")
def train_mnist(encoder_forecaster,
optimizer,
criterion,
lr_scheduler,
batch_size,
max_iterations,
test_iteration_interval,
test_and_save_checkpoint_iterations,
folder_name,
base_dir,
probToPixel=None):
IN_LEN = cfg.MODEL.IN_LEN
OUT_LEN = cfg.MODEL.OUT_LEN
evaluater = HKOEvaluation(seq_len=OUT_LEN, use_central=False)
train_loss = 0.0
save_dir = osp.join(base_dir, folder_name)
if not os.path.exists(save_dir):
os.mkdir(save_dir)
model_save_dir = osp.join(save_dir, 'models')
log_dir = osp.join(save_dir, 'logs')
all_scalars_file_name = osp.join(save_dir, "all_scalars.json")
# pkl_save_dir = osp.join(save_dir, 'pkl')
if osp.exists(all_scalars_file_name):
os.remove(all_scalars_file_name)
if osp.exists(log_dir):
shutil.rmtree(log_dir)
if osp.exists(model_save_dir):
shutil.rmtree(model_save_dir)
os.mkdir(model_save_dir)
writer = SummaryWriter(log_dir)
mnist_iter = MovingMNISTAdvancedIterator(
distractor_num=cfg.MOVINGMNIST.DISTRACTOR_NUM,
initial_velocity_range=(cfg.MOVINGMNIST.VELOCITY_LOWER,
cfg.MOVINGMNIST.VELOCITY_UPPER),
rotation_angle_range=(cfg.MOVINGMNIST.ROTATION_LOWER,
cfg.MOVINGMNIST.ROTATION_UPPER),
scale_variation_range=(cfg.MOVINGMNIST.SCALE_VARIATION_LOWER,
cfg.MOVINGMNIST.SCALE_VARIATION_UPPER),
illumination_factor_range=(cfg.MOVINGMNIST.ILLUMINATION_LOWER,
cfg.MOVINGMNIST.ILLUMINATION_UPPER))
itera = 0
while itera < max_iterations:
frame_dat, _ = mnist_iter.sample(batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
seqlen=cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN)
train_data = torch.from_numpy(
np.array(frame_dat[0:cfg.MOVINGMNIST.IN_LEN, ...])).to(
cfg.GLOBAL.DEVICE) / 255.0
train_label = torch.from_numpy(frame_dat[cfg.MODEL.IN_LEN:(
cfg.MOVINGMNIST.IN_LEN + cfg.MOVINGMNIST.OUT_LEN), ...]).to(
cfg.GLOBAL.DEVICE) / 255.0
encoder_forecaster.train()
optimizer.zero_grad()
output = encoder_forecaster(train_data)
mask = torch.from_numpy(np.ones(train_label.size()).astype(int)).to(
cfg.GLOBAL.DEVICE)
loss = criterion(output, train_label, mask)
loss.backward()
torch.nn.utils.clip_grad_value_(encoder_forecaster.parameters(),
clip_value=50.0)
optimizer.step()
lr_scheduler.step()
train_loss += loss.item()
train_label_numpy = train_label.cpu().numpy()
if probToPixel is None:
output_numpy = np.clip(output.detach().cpu().numpy(), 0.0, 1.0)
else:
# if classification, output: S*B*C*H*W
output_numpy = probToPixel(output.detach().cpu().numpy(),
train_label, mask,
lr_scheduler.get_lr()[0])
evaluater.update(train_label_numpy, output_numpy, mask.cpu().numpy())
if (itera + 1) % test_iteration_interval == 0:
with torch.no_grad():
encoder_forecaster.eval()
overall_mse = 0
for iter_id in range(10):
valid_frame, _ = mnist_iter.sample(
batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
seqlen=cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN,
random=False)
valid_data = torch.from_numpy(
np.array(valid_frame[0:cfg.MOVINGMNIST.IN_LEN,
...])).to(
cfg.GLOBAL.DEVICE) / 255.0
valid_label = torch.from_numpy(
valid_frame[cfg.MODEL.IN_LEN:(cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN),
...]).to(cfg.GLOBAL.DEVICE) / 255.0
output = encoder_forecaster(valid_data)
overall_mse += torch.mean((valid_label - output)**2)
avg_mse = overall_mse / 10
with open(os.path.join(base_dir, 'result.txt'), 'a') as f:
f.write(str(avg_mse) + '\n')
print(base_dir, avg_mse)
gif_dir = os.path.join(base_dir, "gif")
if not os.path.exists(gif_dir):
os.mkdir(gif_dir)
save_gif(output.detach().cpu().numpy()[:, 0, 0, :, :],
os.path.join(gif_dir, "pred-{}.gif".format(itera)))
save_gif(train_data.detach().cpu().numpy()[:, 0, 0, :, :],
os.path.join(gif_dir, "in-{}.gif".format(itera)))
save_gif(train_label.detach().cpu().numpy()[:, 0, 0, :, :],
os.path.join(gif_dir, "gt-{}.gif".format(itera)))
if (itera + 1) % test_and_save_checkpoint_iterations == 0:
torch.save(
encoder_forecaster.state_dict(),
osp.join(model_save_dir,
'encoder_forecaster_{}.pth'.format(itera)))
itera += 1
writer.close()
def train(args):
assert cfg.MODEL.FRAME_STACK == 1 and cfg.MODEL.FRAME_SKIP == 1
base_dir = args.save_dir
logging_config(folder=base_dir, name="training")
save_movingmnist_cfg(base_dir)
mnist_iter = MovingMNISTAdvancedIterator(
distractor_num=cfg.MOVINGMNIST.DISTRACTOR_NUM,
initial_velocity_range=(cfg.MOVINGMNIST.VELOCITY_LOWER,
cfg.MOVINGMNIST.VELOCITY_UPPER),
rotation_angle_range=(cfg.MOVINGMNIST.ROTATION_LOWER,
cfg.MOVINGMNIST.ROTATION_UPPER),
scale_variation_range=(cfg.MOVINGMNIST.SCALE_VARIATION_LOWER,
cfg.MOVINGMNIST.SCALE_VARIATION_UPPER),
illumination_factor_range=(cfg.MOVINGMNIST.ILLUMINATION_LOWER,
cfg.MOVINGMNIST.ILLUMINATION_UPPER))
mnist_rnn = MovingMNISTFactory(batch_size=cfg.MODEL.TRAIN.BATCH_SIZE //
len(args.ctx),
in_seq_len=cfg.MODEL.IN_LEN,
out_seq_len=cfg.MODEL.OUT_LEN)
encoder_net, forecaster_net, loss_net = \
encoder_forecaster_build_networks(
factory=mnist_rnn,
context=args.ctx)
t_encoder_net, t_forecaster_net, t_loss_net = \
encoder_forecaster_build_networks(
factory=mnist_rnn,
context=args.ctx[0],
shared_encoder_net=encoder_net,
shared_forecaster_net=forecaster_net,
shared_loss_net=loss_net,
for_finetune=True)
encoder_net.summary()
forecaster_net.summary()
loss_net.summary()
# Resume last checkpoint
if args.resume:
encoder_net.load(prefix=os.path.join(base_dir, 'encoder_net'),
epoch=latest_iter_id(base_dir),
load_optimizer_states=True,
data_names=[
'data', 'ebrnn1_begin_state_h',
'ebrnn2_begin_state_h', 'ebrnn3_begin_state_h'
],
label_names=[]) # change it next time
forecaster_net.load(prefix=os.path.join(base_dir, 'forecaster_net'),
epoch=latest_iter_id(base_dir),
load_optimizer_states=True,
data_names=[
'fbrnn1_begin_state_h', 'fbrnn2_begin_state_h',
'fbrnn3_begin_state_h'
],
label_names=[]) # change it next time
# Begin to load the model if load_dir is not empty
if len(cfg.MODEL.LOAD_DIR) > 0:
load_mnist_params(load_dir=cfg.MODEL.LOAD_DIR,
load_iter=cfg.MODEL.LOAD_ITER,
encoder_net=encoder_net,
forecaster_net=forecaster_net)
states = EncoderForecasterStates(factory=mnist_rnn, ctx=args.ctx[0])
states.reset_all()
for info in mnist_rnn.init_encoder_state_info:
assert info["__layout__"].find(
'N') == 0, "Layout=%s is not supported!" % info["__layout__"]
iter_id = 0
while iter_id < cfg.MODEL.TRAIN.MAX_ITER:
frame_dat, _ = mnist_iter.sample(batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
seqlen=cfg.MOVINGMNIST.IN_LEN +
cfg.MOVINGMNIST.OUT_LEN)
data_nd = mx.nd.array(frame_dat[0:cfg.MOVINGMNIST.IN_LEN, ...],
ctx=args.ctx[0]) / 255.0
target_nd = mx.nd.array(frame_dat[cfg.MODEL.IN_LEN:(
cfg.MOVINGMNIST.IN_LEN + cfg.MOVINGMNIST.OUT_LEN), ...],
ctx=args.ctx[0]) / 255.0
train_step(batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
encoder_net=encoder_net,
forecaster_net=forecaster_net,
loss_net=loss_net,
init_states=states,
data_nd=data_nd,
gt_nd=target_nd,
mask_nd=None,
iter_id=iter_id)
if (iter_id + 1) % 100 == 0:
new_frame_dat, _ = mnist_iter.sample(
batch_size=cfg.MODEL.TRAIN.BATCH_SIZE,
seqlen=cfg.MOVINGMNIST.IN_LEN + cfg.MOVINGMNIST.OUT_LEN)
data_nd = mx.nd.array(frame_dat[0:cfg.MOVINGMNIST.IN_LEN, ...],
ctx=args.ctx[0]) / 255.0
target_nd = mx.nd.array(frame_dat[cfg.MOVINGMNIST.IN_LEN:(
cfg.MOVINGMNIST.IN_LEN + cfg.MOVINGMNIST.OUT_LEN), ...],
ctx=args.ctx[0]) / 255.0
pred_nd = mnist_get_prediction(data_nd=data_nd,
states=states,
encoder_net=encoder_net,
forecaster_net=forecaster_net)
save_gif(pred_nd.asnumpy()[:, 0, 0, :, :],
os.path.join(base_dir, "pred.gif"))
save_gif(data_nd.asnumpy()[:, 0, 0, :, :],
os.path.join(base_dir, "in.gif"))
save_gif(target_nd.asnumpy()[:, 0, 0, :, :],
os.path.join(base_dir, "gt.gif"))
if (iter_id + 1) % cfg.MODEL.SAVE_ITER == 0:
encoder_net.save_checkpoint(prefix=os.path.join(
base_dir, "encoder_net"),
epoch=iter_id,
save_optimizer_states=True)
forecaster_net.save_checkpoint(prefix=os.path.join(
base_dir, "forecaster_net"),
epoch=iter_id,
save_optimizer_states=True)
iter_id += 1