Update anime_aesthetic.py
Browse files- anime_aesthetic.py +10 -3
anime_aesthetic.py
CHANGED
@@ -216,11 +216,18 @@ def rescale_pad(image, output_size, random_pad=False):
|
|
216 |
if h != output_size or w != output_size:
|
217 |
r = min(output_size / h, output_size / w)
|
218 |
new_h, new_w = int(h * r), int(w * r)
|
|
|
|
|
|
|
219 |
ph = output_size - new_h
|
220 |
pw = output_size - new_w
|
|
|
|
|
|
|
|
|
221 |
image = transforms.functional.resize(image, [new_h, new_w])
|
222 |
image = transforms.functional.pad(
|
223 |
-
image, [
|
224 |
)
|
225 |
return image
|
226 |
|
@@ -435,7 +442,7 @@ if __name__ == "__main__":
|
|
435 |
parser.add_argument(
|
436 |
"--data-split",
|
437 |
type=float,
|
438 |
-
default=0.
|
439 |
help="split rate for training and validation",
|
440 |
)
|
441 |
|
@@ -486,7 +493,7 @@ if __name__ == "__main__":
|
|
486 |
"--log-step", type=int, default=2, help="log training loss every n steps"
|
487 |
)
|
488 |
parser.add_argument(
|
489 |
-
"--val-epoch", type=int, default=0.
|
490 |
)
|
491 |
|
492 |
opt = parser.parse_args()
|
|
|
216 |
if h != output_size or w != output_size:
|
217 |
r = min(output_size / h, output_size / w)
|
218 |
new_h, new_w = int(h * r), int(w * r)
|
219 |
+
if random_pad:
|
220 |
+
r2 = random.uniform(0.9, 1)
|
221 |
+
new_h, new_w = int(new_h * r2), int(new_w * r2)
|
222 |
ph = output_size - new_h
|
223 |
pw = output_size - new_w
|
224 |
+
left = random.randint(0, pw) if random_pad else pw // 2
|
225 |
+
right = pw - left
|
226 |
+
top = random.randint(0, ph) if random_pad else ph // 2
|
227 |
+
bottom = ph - top
|
228 |
image = transforms.functional.resize(image, [new_h, new_w])
|
229 |
image = transforms.functional.pad(
|
230 |
+
image, [left, top, right, bottom], random.uniform(0, 1) if random_pad else 0
|
231 |
)
|
232 |
return image
|
233 |
|
|
|
442 |
parser.add_argument(
|
443 |
"--data-split",
|
444 |
type=float,
|
445 |
+
default=0.9999,
|
446 |
help="split rate for training and validation",
|
447 |
)
|
448 |
|
|
|
493 |
"--log-step", type=int, default=2, help="log training loss every n steps"
|
494 |
)
|
495 |
parser.add_argument(
|
496 |
+
"--val-epoch", type=int, default=0.025, help="valid and save every n epoch"
|
497 |
)
|
498 |
|
499 |
opt = parser.parse_args()
|