def run(self):
self.model.train()
for epoch in range(self.EPOCH_START, self.EPOCHS):
self.epoch = epoch
self.epoch_ts = time.time()
self.running_loss = 0.0
utils.iterate_loader(self.DEVICE, self.loaders["train"],
self.__train_step_fn)
if self.SHOULD_CHECKPOINT:
checkpoint_file = f"epoch_{epoch+1}.pt"
path = os.path.join(self.checkpoint_dir, checkpoint_file)
print(f"Saving checkpoint: {path}")
torch.save(
{
"epoch": epoch,
"model": self.model.state_dict(),
"optimizer": self.optimizer.state_dict()
}, path)
if self.SHOULD_WRITE:
self.writer.close()
def __validate(self):
N = len(self.loaders["val"])
val_metrics = {}
train_metrics = {}
utils.iterate_loader(self.DEVICE,
self.loaders["val"],
self.__val_step_fn,
args=(val_metrics, ))
utils.iterate_loader(self.DEVICE,
self.loaders["train"],
self.__val_step_fn,
args=(train_metrics, ),
end=N)
val_metrics = utils.map_dict(val_metrics, lambda v: v / N)
train_metrics = utils.map_dict(train_metrics, lambda v: v / N)
return val_metrics, train_metrics
def run(self):
self.model.eval()
utils.iterate_loader(self.DEVICE,
self.loader,
self._step_fn,
end=len(self.loader))
mean_ates = np.mean(self.ates)
std_ates = np.std(self.ates)
mean_metrics = utils.dict_mean(self.metrics)
std_metrics = utils.dict_std(self.metrics)
print("\n" + "=" * 20)
print(f"Trajectory error: {mean_ates:0.3f}, std: {std_ates:0.3f}")
print("Depth metrics:")
for key in self.metrics:
print(
f"{key} -> mean: {mean_metrics[key]:0.3f}, std: {std_metrics[key]:0.3f}"
)
print("=" * 20)
def main():
# Parse arguments
args = options.get_args(description="Debug a network",
options=[
"batch",
"workers",
"device",
"load",
"net",
"loss",
])
if args.load == "":
print("No model file specified to load!")
exit()
# Construct datasets
random.seed(1337)
_, _, test_loader = utils.get_kitti_split(args.batch, args.workers)
# The model
model = networks.architectures.get_net(args)
loss_fn = sfm_loss.get_loss_fn(args)
# Load
checkpoint = torch.load(args.load, map_location=torch.device(args.device))
model.load_state_dict(checkpoint["model"])
# Window
pango.CreateWindowAndBind('Main', int(640 * (3 / 2)), int(480 * (3 / 2)))
gl.glEnable(gl.GL_DEPTH_TEST)
# Define Projection and initial ModelView matrix
scam = pango.OpenGlRenderState(
pango.ProjectionMatrix(640, 480, 420, 420, 320, 240, 0.2, 200),
pango.ModelViewLookAt(0, -0.5, -0.5, 0, 0, 1, 0, -1, 0))
handler = pango.Handler3D(scam)
# Create Interactive View in window
dcam = pango.CreateDisplay()
dcam.SetBounds(pango.Attach(0), pango.Attach(1), pango.Attach(0),
pango.Attach(1), -640.0 / 480.0)
dcam.SetHandler(handler)
# Test
model.train()
model = model.to(args.device)
def step_fn(step, inputs):
# Forward pass and loss
with torch.no_grad():
loss, data = utils.forward_pass(model, loss_fn, inputs)
print("loss %f" % loss.item())
print(data.keys())
print(data["pose"].shape)
for i in range(args.batch):
print(list(data["pose"][i, 0, :].cpu().detach().numpy()))
print(list(data["pose"][i, 1, :].cpu().detach().numpy()))
print("--")
depth_img = viz.tensor2depthimg(
torch.cat((*data["depth"][0][:, 0], ), dim=0))
tgt_img = viz.tensor2img(torch.cat((*data["tgt"], ), dim=1))
img = np.concatenate((tgt_img, depth_img), axis=1)
warp_imgs = []
#diff_imgs = []
for warp, diff in zip(data["warp"], data["diff"]):
warp = restack(restack(warp, 1, -1), 0, -2)
diff = restack(restack(diff, 1, -1), 0, -2)
warp_imgs.append(viz.tensor2img(warp))
#diff_imgs.append(viz.tensor2diffimg(diff))
world = reconstruction.depth_to_3d_points(data["depth"][0], data["K"])
points = world[0, :].view(3, -1).transpose(
1, 0).cpu().detach().numpy().astype(np.float64)
colors = (data["tgt"][0, :].view(3, -1).transpose(
1, 0).cpu().detach().numpy().astype(np.float64) + 1) / 2
loop = True
while loop:
key = cv2.waitKey(10)
if key == 27 or pango.ShouldQuit():
exit()
elif key != -1:
loop = False
cv2.imshow("target and depth", img)
#for i, (warp, diff) in enumerate(zip(warp_imgs, diff_imgs)):
for i, warp in enumerate(warp_imgs):
cv2.imshow("warp scale: %d" % i, warp)
#cv2.imshow("diff scale: %d" % i, diff)
gl.glClear(gl.GL_COLOR_BUFFER_BIT | gl.GL_DEPTH_BUFFER_BIT)
gl.glClearColor(1.0, 1.0, 1.0, 1.0)
dcam.Activate(scam)
gl.glPointSize(5)
pango.DrawPoints(points, colors)
pose = np.identity(4)
pose[:3, 3] = 0
gl.glLineWidth(1)
gl.glColor3f(0.0, 0.0, 1.0)
pango.DrawCamera(pose, 0.5, 0.75, 0.8)
pango.FinishFrame()
utils.iterate_loader(args, test_loader, step_fn)
def main():
# Parse arguments
args = options.get_args(description="Debug a network",
options=[
"batch",
"workers",
"device",
"load",
"smooth-weight",
"explain-weight",
"net",
])
if args.load == "":
print("No model file specified to load!")
exit()
# Construct datasets
random.seed(1337)
_, _, test_loader = utils.get_kitti_split(args.batch, args.workers)
# The model
model = networks.architectures.get_net(args)
loss_fn = sfm_loss.get_loss_fn(args)
# Load
checkpoint = torch.load(args.load, map_location=torch.device(args.device))
model.load_state_dict(checkpoint["model"])
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1000, 500))
# Test
model.train()
model = model.to(args.device)
def step_fn(step, inputs):
# Forward pass and loss
with torch.no_grad():
loss, data = utils.forward_pass(model, loss_fn, inputs)
print("loss %f" % loss.item())
print(data.keys())
print(data["pose"].shape)
for i in range(4):
print(list(data["pose"][i, 0, :].cpu().detach().numpy()))
print(list(data["pose"][i, 1, :].cpu().detach().numpy()))
print("--")
depth_img = viz.tensor2depthimg(
torch.cat((*data["depth"][0][:, 0], ), dim=0))
tgt_img = viz.tensor2img(torch.cat((*data["tgt"], ), dim=1))
img = np.concatenate((tgt_img, depth_img), axis=1)
warp_imgs = []
diff_imgs = []
for warp, diff in zip(data["warp"], data["diff"]):
warp = restack(restack(warp, 1, -1), 0, -2)
diff = restack(restack(diff, 1, -1), 0, -2)
warp_imgs.append(viz.tensor2img(warp))
diff_imgs.append(viz.tensor2diffimg(diff))
world = inverse_warp.depth_to_3d_points(data["depth"][0], data["K"])
points = world[0, :].view(3, -1).transpose(
1, 0).cpu().detach().numpy().astype(np.float64)
colors = (data["tgt"][0, :].view(3, -1).transpose(
1, 0).cpu().detach().numpy().astype(np.float64) + 1) / 2
test_mayavi.draw_rgb_points(fig, points, colors)
loop = True
while loop:
key = cv2.waitKey(10)
if key == 27:
exit()
elif key != -1:
loop = False
cv2.imshow("target and depth", img)
for i, (warp, diff) in enumerate(zip(warp_imgs, diff_imgs)):
cv2.imshow("warp scale: %d" % i, warp)
cv2.imshow("diff scale: %d" % i, diff)
mlab.show(10)
utils.iterate_loader(args, test_loader, step_fn)
def run(self):
self.model.eval()
utils.iterate_loader(self.DEVICE, self.loader, self._step_fn)