yaron123 commited on
Commit
7f44c6b
·
1 Parent(s): 6b44be4
Files changed (1) hide show
  1. app.py +25 -26
app.py CHANGED
@@ -413,7 +413,8 @@ CHECKPOINTS = ESRGANUpscalerCheckpoints(
413
 
414
  device = DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
415
  DTYPE = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
416
- enhancer = ESRGANUpscaler(checkpoints=CHECKPOINTS, device="cpu", dtype=DTYPE)
 
417
 
418
  # logging
419
 
@@ -434,8 +435,8 @@ pegasus_name = "google/pegasus-xsum"
434
  # precision data
435
 
436
  seq=512
437
- width=1500
438
- height=1500
439
  image_steps=8
440
  img_accu=0
441
 
@@ -554,7 +555,7 @@ def upscaler(
554
  return enhanced_image
555
 
556
  def summarize_text(
557
- text, max_length=30, num_beams=4, early_stopping=True
558
  ):
559
  log(f'CALL summarize_text')
560
  summary = pegasus_tokenizer.decode( pegasus_model.generate(
@@ -570,8 +571,7 @@ def generate_random_string(length):
570
  characters = str(ascii_letters + digits)
571
  return ''.join(random.choice(characters) for _ in range(length))
572
 
573
- @spaces.GPU(duration=180)
574
- def pipe_generate(p1,p2):
575
  log(f'CALL pipe_generate')
576
  imgs = image_pipe(
577
  prompt=p1,
@@ -579,7 +579,7 @@ def pipe_generate(p1,p2):
579
  height=height,
580
  width=width,
581
  guidance_scale=img_accu,
582
- num_images_per_prompt=6,
583
  num_inference_steps=image_steps,
584
  max_sequence_length=seq,
585
  generator=torch.Generator(device).manual_seed(int(str(random.random()).split(".")[1]))
@@ -612,6 +612,20 @@ def add_song_cover_text(img,artist,song):
612
 
613
  return img
614
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
615
  def handle_generate(artist,song,genre,lyrics):
616
 
617
  log(f'CALL handle_generate')
@@ -631,19 +645,9 @@ def handle_generate(artist,song,genre,lyrics):
631
 
632
  Negative: {neg}
633
  """)
634
-
635
- imgs = pipe_generate(pos,neg)
636
-
637
- names = []
638
- index = 1
639
- for img in imgs:
640
- labeled_img = add_song_cover_text(img,pos_artist,pos_song)
641
- enhanced_img = upscaler(labeled_img)
642
- name = f'{pos_artist} - {pos_song} ({index}).png'
643
- enhanced_img.save(name)
644
- names.append(name)
645
- return names
646
 
 
 
647
  # entry
648
 
649
  if __name__ == "__main__":
@@ -682,17 +686,12 @@ if __name__ == "__main__":
682
  run = gr.Button("Generate",elem_classes="btn")
683
 
684
  with gr.Row():
685
- cover1 = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
686
- cover2 = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
687
- cover3 = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
688
- cover4 = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
689
- cover5 = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
690
- cover6 = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
691
 
692
  run.click(
693
  fn=handle_generate,
694
  inputs=[artist,song,genre,lyrics],
695
- outputs=[cover1,cover2,cover3,cover4,cover5,cover6]
696
  )
697
 
698
  demo.queue().launch()
 
413
 
414
  device = DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
415
  DTYPE = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
416
+ enhancer = ESRGANUpscaler(checkpoints=CHECKPOINTS, device=DEVICE, dtype=DTYPE)
417
+ enhancer.to(DEVICE)
418
 
419
  # logging
420
 
 
435
  # precision data
436
 
437
  seq=512
438
+ width=1536
439
+ height=1536
440
  image_steps=8
441
  img_accu=0
442
 
 
555
  return enhanced_image
556
 
557
  def summarize_text(
558
+ text, max_length=10, num_beams=4, early_stopping=True
559
  ):
560
  log(f'CALL summarize_text')
561
  summary = pegasus_tokenizer.decode( pegasus_model.generate(
 
571
  characters = str(ascii_letters + digits)
572
  return ''.join(random.choice(characters) for _ in range(length))
573
 
574
+ def pipe_generate_image(p1,p2):
 
575
  log(f'CALL pipe_generate')
576
  imgs = image_pipe(
577
  prompt=p1,
 
579
  height=height,
580
  width=width,
581
  guidance_scale=img_accu,
582
+ num_images_per_prompt=1,
583
  num_inference_steps=image_steps,
584
  max_sequence_length=seq,
585
  generator=torch.Generator(device).manual_seed(int(str(random.random()).split(".")[1]))
 
612
 
613
  return img
614
 
615
+ @spaces.GPU(duration=180)
616
+ def all_pipes(img,pos,neg,artist,song):
617
+ imgs = pipe_generate_image(pos,neg)
618
+
619
+ names = []
620
+ index = 1
621
+ for img in imgs:
622
+ labeled_img = add_song_cover_text(img,artist,song)
623
+ enhanced_img = upscaler(labeled_img)
624
+ name = f'{pos_artist} - {pos_song} ({index}).png'
625
+ enhanced_img.save(name)
626
+ names.append(name)
627
+ return names
628
+
629
  def handle_generate(artist,song,genre,lyrics):
630
 
631
  log(f'CALL handle_generate')
 
645
 
646
  Negative: {neg}
647
  """)
 
 
 
 
 
 
 
 
 
 
 
 
648
 
649
+ return all_pipes(img,pos,neg,pos_artist,pos_song)
650
+
651
  # entry
652
 
653
  if __name__ == "__main__":
 
686
  run = gr.Button("Generate",elem_classes="btn")
687
 
688
  with gr.Row():
689
+ cover = gr.Image(interactive=False,container=False,elem_classes="image-container", label="Result", show_label=True, type='filepath', show_share_button=False)
 
 
 
 
 
690
 
691
  run.click(
692
  fn=handle_generate,
693
  inputs=[artist,song,genre,lyrics],
694
+ outputs=[cover]
695
  )
696
 
697
  demo.queue().launch()