def main():
print(DATA_DIR)
plif_dataset = PLIF(plif_dir=DATA_DIR)
plif_dataloader = torch.utils.data.DataLoader(plif_dataset,
batch_size=1,
pin_memory=PIN_MEMORY,
num_workers=1)
img_height, img_width = plif_dataset[0].squeeze(0).numpy().shape
with torch.no_grad():
model = DifferenceOfGaussiansStandardConv(
min_sigma=1,
max_sigma=10,
overlap=0.9,
threshold=0.012,
prune=False,
sigma_bins=20,
).to(DEVICE, non_blocking=PIN_MEMORY)
for p in model.parameters():
p.requires_grad = False
model.eval()
start = time.monotonic()
for i, blobs in enumerate(torch_dog(plif_dataloader, model)):
print(time.monotonic() - start)
start = time.monotonic()
def main(data_dir_path):
plif_dataset = PLIF(plif_dir=Path(data_dir_path))
plif_dataloader = torch.utils.data.DataLoader(plif_dataset,
batch_size=1,
pin_memory=PIN_MEMORY,
num_workers=1)
img_height, img_width = plif_dataset[0].squeeze(0).numpy().shape
with torch.no_grad():
model = DifferenceOfGaussiansStandardConv(
min_sigma=1,
max_sigma=10,
overlap=0.9,
threshold=0.012,
prune=False,
sigma_bins=20,
).to(DEVICE, non_blocking=PIN_MEMORY)
for p in model.parameters():
p.requires_grad = False
model.eval()
for i, img_tensor in enumerate(plif_dataloader):
img_tensor = img_tensor.to(DEVICE, non_blocking=PIN_MEMORY)
mask, local_maxima = model(img_tensor.unsqueeze(0))
blobs = model.make_blobs(mask, local_maxima)
make_circles_fig(plif_dataset[0], blobs).show()
break
def gpu_standard_run_times(fn):
image_pth = Path(os.path.dirname(os.path.realpath(__file__))) / Path(
"../simulation/screenshot.png"
)
screenshot = SimulPLIF(img_path=image_pth, num_repeats=1000000)
train_dataloader = torch.utils.data.DataLoader(
screenshot, batch_size=1, pin_memory=PIN_MEMORY
)
with open(f"{fn}.csv", "w", newline="") as csvfile:
writer = csv.writer(csvfile)
writer.writerow(["n_bin", "max_sigma", "time"])
for max_sigma in range(min_sigma + 1, mx_sigma + 1):
for n_bin in range(min_bin, max_bin + 1):
with torch.no_grad():
model = DifferenceOfGaussiansStandardConv(
min_sigma=min_sigma,
max_sigma=max_sigma,
overlap=0.9,
threshold=0.012,
prune=False,
sigma_bins=n_bin,
).to(DEVICE, non_blocking=PIN_MEMORY)
for p in model.parameters():
p.requires_grad = False
model.eval()
for i in range(REPEATS):
img_tensor = next(iter(train_dataloader))
img_tensor = img_tensor.to(DEVICE, non_blocking=PIN_MEMORY)
torch.cuda.synchronize(DEVICE)
START = time.monotonic()
mask, local_maxima = model(img_tensor.unsqueeze(0))
blobs = model.make_blobs(mask, local_maxima)
torch.cuda.synchronize(DEVICE)
END = time.monotonic()
res = [n_bin, max_sigma, END - START]
writer.writerow(res)
print(res)
del model
def test_simul():
image_pth = Path(os.path.dirname(
os.path.realpath(__file__))) / Path("../simulation/screenshot.png")
screenshot = SimulPLIF(img_path=image_pth, num_repeats=1000000)
with torch.no_grad():
model = DifferenceOfGaussiansStandardConv(
min_sigma=1,
max_sigma=10,
overlap=0.9,
# threshold=0.012,
threshold=0.112,
prune=False,
sigma_bins=20,
).to(DEVICE, non_blocking=PIN_MEMORY)
for p in model.parameters():
p.requires_grad = False
model.eval()
img_tensor = screenshot[0]
img_tensor = img_tensor.to(DEVICE, non_blocking=PIN_MEMORY)
mask, local_maxima = model(img_tensor.unsqueeze(0).unsqueeze(0))
blobs = model.make_blobs(mask, local_maxima)
make_circles_fig(screenshot[0], blobs).show()