Spaces:
Paused
Paused
alessandro trinca tornidor
commited on
Commit
·
37a5f04
1
Parent(s):
e789db0
[refactor] set image precision with an external function
Browse files
app.py
CHANGED
@@ -100,6 +100,16 @@ def parse_args(args_to_parse):
|
|
100 |
return parser.parse_args(args_to_parse)
|
101 |
|
102 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
def preprocess(
|
104 |
x,
|
105 |
pixel_mean=torch.Tensor([123.675, 116.28, 103.53]).view(-1, 1, 1),
|
@@ -267,12 +277,7 @@ def get_inference_model_by_args(args_to_parse):
|
|
267 |
.cuda()
|
268 |
)
|
269 |
logging.info(f"image_clip type: {type(image_clip)}.")
|
270 |
-
|
271 |
-
image_clip = image_clip.bfloat16()
|
272 |
-
elif args_to_parse.precision == "fp16":
|
273 |
-
image_clip = image_clip.half()
|
274 |
-
else:
|
275 |
-
image_clip = image_clip.float()
|
276 |
|
277 |
image = transform.apply_image(image_np)
|
278 |
resize_list = [image.shape[:2]]
|
@@ -283,12 +288,7 @@ def get_inference_model_by_args(args_to_parse):
|
|
283 |
.cuda()
|
284 |
)
|
285 |
logging.info(f"image_clip type: {type(image_clip)}.")
|
286 |
-
|
287 |
-
image = image.bfloat16()
|
288 |
-
elif args_to_parse.precision == "fp16":
|
289 |
-
image = image.half()
|
290 |
-
else:
|
291 |
-
image = image.float()
|
292 |
|
293 |
input_ids = tokenizer_image_token(prompt, tokenizer, return_tensors="pt")
|
294 |
input_ids = input_ids.unsqueeze(0).cuda()
|
@@ -330,6 +330,7 @@ def get_inference_model_by_args(args_to_parse):
|
|
330 |
## no seg output
|
331 |
output_image = cv2.imread("./resources/no_seg_out.png")[:, :, ::-1]
|
332 |
return output_image, output_str
|
|
|
333 |
return inference
|
334 |
|
335 |
|
|
|
100 |
return parser.parse_args(args_to_parse)
|
101 |
|
102 |
|
103 |
+
def set_image_precision_by_args(input_image, precision):
|
104 |
+
if precision == "bf16":
|
105 |
+
input_image = input_image.bfloat16()
|
106 |
+
elif precision == "fp16":
|
107 |
+
input_image = input_image.half()
|
108 |
+
else:
|
109 |
+
input_image = input_image.float()
|
110 |
+
return input_image
|
111 |
+
|
112 |
+
|
113 |
def preprocess(
|
114 |
x,
|
115 |
pixel_mean=torch.Tensor([123.675, 116.28, 103.53]).view(-1, 1, 1),
|
|
|
277 |
.cuda()
|
278 |
)
|
279 |
logging.info(f"image_clip type: {type(image_clip)}.")
|
280 |
+
image_clip = set_image_precision_by_args(image_clip, args_to_parse.precision)
|
|
|
|
|
|
|
|
|
|
|
281 |
|
282 |
image = transform.apply_image(image_np)
|
283 |
resize_list = [image.shape[:2]]
|
|
|
288 |
.cuda()
|
289 |
)
|
290 |
logging.info(f"image_clip type: {type(image_clip)}.")
|
291 |
+
image = set_image_precision_by_args(image, args_to_parse.precision)
|
|
|
|
|
|
|
|
|
|
|
292 |
|
293 |
input_ids = tokenizer_image_token(prompt, tokenizer, return_tensors="pt")
|
294 |
input_ids = input_ids.unsqueeze(0).cuda()
|
|
|
330 |
## no seg output
|
331 |
output_image = cv2.imread("./resources/no_seg_out.png")[:, :, ::-1]
|
332 |
return output_image, output_str
|
333 |
+
|
334 |
return inference
|
335 |
|
336 |
|