remove sam2
Browse files- app.py +101 -54
- requirements.txt +1 -1
app.py
CHANGED
@@ -3,11 +3,11 @@ import spaces
|
|
3 |
import torch
|
4 |
from loadimg import load_img
|
5 |
from torchvision import transforms
|
6 |
-
from transformers import AutoModelForImageSegmentation
|
7 |
from diffusers import FluxFillPipeline
|
8 |
from PIL import Image, ImageOps
|
9 |
|
10 |
-
from sam2.sam2_image_predictor import SAM2ImagePredictor
|
11 |
import numpy as np
|
12 |
from simple_lama_inpainting import SimpleLama
|
13 |
from contextlib import contextmanager
|
@@ -134,36 +134,36 @@ def rmbg(image=None, url=None):
|
|
134 |
return image
|
135 |
|
136 |
|
137 |
-
def mask_generation(image=None, d=None):
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
|
168 |
|
169 |
def erase(image=None, mask=None):
|
@@ -173,6 +173,25 @@ def erase(image=None, mask=None):
|
|
173 |
return simple_lama(image, mask)
|
174 |
|
175 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
176 |
@spaces.GPU(duration=120)
|
177 |
def main(*args):
|
178 |
api_num = args[0]
|
@@ -183,10 +202,12 @@ def main(*args):
|
|
183 |
return outpaint(*args)
|
184 |
elif api_num == 3:
|
185 |
return inpaint(*args)
|
186 |
-
elif api_num == 4:
|
187 |
-
|
188 |
elif api_num == 5:
|
189 |
return erase(*args)
|
|
|
|
|
190 |
|
191 |
|
192 |
rmbg_tab = gr.Interface(
|
@@ -241,24 +262,24 @@ inpaint_tab = gr.Interface(
|
|
241 |
)
|
242 |
|
243 |
|
244 |
-
sam2_tab = gr.Interface(
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
)
|
262 |
|
263 |
erase_tab = gr.Interface(
|
264 |
main,
|
@@ -279,9 +300,35 @@ erase_tab = gr.Interface(
|
|
279 |
cache_examples=False,
|
280 |
)
|
281 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
282 |
demo = gr.TabbedInterface(
|
283 |
-
[
|
284 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
285 |
title="Utilities that require GPU",
|
286 |
)
|
287 |
|
|
|
3 |
import torch
|
4 |
from loadimg import load_img
|
5 |
from torchvision import transforms
|
6 |
+
from transformers import AutoModelForImageSegmentation, pipeline
|
7 |
from diffusers import FluxFillPipeline
|
8 |
from PIL import Image, ImageOps
|
9 |
|
10 |
+
# from sam2.sam2_image_predictor import SAM2ImagePredictor
|
11 |
import numpy as np
|
12 |
from simple_lama_inpainting import SimpleLama
|
13 |
from contextlib import contextmanager
|
|
|
134 |
return image
|
135 |
|
136 |
|
137 |
+
# def mask_generation(image=None, d=None):
|
138 |
+
# # use bfloat16 for the entire notebook
|
139 |
+
# # torch.autocast("cuda", dtype=torch.bfloat16).__enter__()
|
140 |
+
# # # turn on tfloat32 for Ampere GPUs (https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices)
|
141 |
+
# # if torch.cuda.get_device_properties(0).major >= 8:
|
142 |
+
# # torch.backends.cuda.matmul.allow_tf32 = True
|
143 |
+
# # torch.backends.cudnn.allow_tf32 = True
|
144 |
+
# d = eval(d) # convert this to dictionary
|
145 |
+
# with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
|
146 |
+
# predictor = SAM2ImagePredictor.from_pretrained("facebook/sam2.1-hiera-large")
|
147 |
+
# predictor.set_image(image)
|
148 |
+
# input_point = np.array(d["input_points"])
|
149 |
+
# input_label = np.array(d["input_labels"])
|
150 |
+
# masks, scores, logits = predictor.predict(
|
151 |
+
# point_coords=input_point,
|
152 |
+
# point_labels=input_label,
|
153 |
+
# multimask_output=True,
|
154 |
+
# )
|
155 |
+
# sorted_ind = np.argsort(scores)[::-1]
|
156 |
+
# masks = masks[sorted_ind]
|
157 |
+
# scores = scores[sorted_ind]
|
158 |
+
# logits = logits[sorted_ind]
|
159 |
+
|
160 |
+
# out = []
|
161 |
+
# for i in range(len(masks)):
|
162 |
+
# m = Image.fromarray(masks[i] * 255).convert("L")
|
163 |
+
# comp = Image.composite(image, m, m)
|
164 |
+
# out.append((comp, f"image {i}"))
|
165 |
+
|
166 |
+
# return out
|
167 |
|
168 |
|
169 |
def erase(image=None, mask=None):
|
|
|
173 |
return simple_lama(image, mask)
|
174 |
|
175 |
|
176 |
+
# Initialize Whisper model
|
177 |
+
whisper = pipeline(
|
178 |
+
task="automatic-speech-recognition",
|
179 |
+
model="openai/whisper-large-v3",
|
180 |
+
chunk_length_s=30,
|
181 |
+
device="cuda" if torch.cuda.is_available() else "cpu",
|
182 |
+
)
|
183 |
+
|
184 |
+
|
185 |
+
def transcribe(audio, task="transcribe"):
|
186 |
+
if audio is None:
|
187 |
+
raise gr.Error("No audio file submitted!")
|
188 |
+
|
189 |
+
text = whisper(
|
190 |
+
audio, batch_size=8, generate_kwargs={"task": task}, return_timestamps=True
|
191 |
+
)["text"]
|
192 |
+
return text
|
193 |
+
|
194 |
+
|
195 |
@spaces.GPU(duration=120)
|
196 |
def main(*args):
|
197 |
api_num = args[0]
|
|
|
202 |
return outpaint(*args)
|
203 |
elif api_num == 3:
|
204 |
return inpaint(*args)
|
205 |
+
# elif api_num == 4:
|
206 |
+
# return mask_generation(*args)
|
207 |
elif api_num == 5:
|
208 |
return erase(*args)
|
209 |
+
elif api_num == 6:
|
210 |
+
return transcribe(*args)
|
211 |
|
212 |
|
213 |
rmbg_tab = gr.Interface(
|
|
|
262 |
)
|
263 |
|
264 |
|
265 |
+
# sam2_tab = gr.Interface(
|
266 |
+
# main,
|
267 |
+
# inputs=[
|
268 |
+
# gr.Number(4, interactive=False),
|
269 |
+
# gr.Image(type="pil"),
|
270 |
+
# gr.Text(),
|
271 |
+
# ],
|
272 |
+
# outputs=gr.Gallery(),
|
273 |
+
# examples=[
|
274 |
+
# [
|
275 |
+
# 4,
|
276 |
+
# "./assets/truck.jpg",
|
277 |
+
# '{"input_points": [[500, 375], [1125, 625]], "input_labels": [1, 0]}',
|
278 |
+
# ]
|
279 |
+
# ],
|
280 |
+
# api_name="sam2",
|
281 |
+
# cache_examples=False,
|
282 |
+
# )
|
283 |
|
284 |
erase_tab = gr.Interface(
|
285 |
main,
|
|
|
300 |
cache_examples=False,
|
301 |
)
|
302 |
|
303 |
+
transcribe_tab = gr.Interface(
|
304 |
+
fn=main,
|
305 |
+
inputs=[
|
306 |
+
gr.Number(6, interactive=False),
|
307 |
+
gr.Audio(source="upload", type="filepath"),
|
308 |
+
gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
|
309 |
+
],
|
310 |
+
outputs="text",
|
311 |
+
api_name="transcribe",
|
312 |
+
description="Upload an audio file to extract text using Whisper Large V3",
|
313 |
+
)
|
314 |
+
|
315 |
demo = gr.TabbedInterface(
|
316 |
+
[
|
317 |
+
rmbg_tab,
|
318 |
+
outpaint_tab,
|
319 |
+
inpaint_tab,
|
320 |
+
# sam2_tab,
|
321 |
+
erase_tab,
|
322 |
+
transcribe_tab,
|
323 |
+
],
|
324 |
+
[
|
325 |
+
"remove background",
|
326 |
+
"outpainting",
|
327 |
+
"inpainting",
|
328 |
+
# "sam2",
|
329 |
+
"erase",
|
330 |
+
"transcribe",
|
331 |
+
],
|
332 |
title="Utilities that require GPU",
|
333 |
)
|
334 |
|
requirements.txt
CHANGED
@@ -19,6 +19,6 @@ kornia
|
|
19 |
huggingface_hub
|
20 |
sentencepiece
|
21 |
einops
|
22 |
-
git+https://github.com/facebookresearch/sam2.git
|
23 |
matplotlib
|
24 |
simple-lama-inpainting
|
|
|
19 |
huggingface_hub
|
20 |
sentencepiece
|
21 |
einops
|
22 |
+
# git+https://github.com/facebookresearch/sam2.git
|
23 |
matplotlib
|
24 |
simple-lama-inpainting
|