tcm03 commited on
Commit
fc48d96
·
1 Parent(s): cc6eb9f

Analyze running time of process_images() and process_video_frames()

Browse files
Files changed (1) hide show
  1. preprocessing/mm_datautils.py +8 -5
preprocessing/mm_datautils.py CHANGED
@@ -4,6 +4,7 @@ import torch
4
  from decord import cpu, VideoReader
5
  from transformers import BaseImageProcessor
6
  from typing import List, Union, Tuple
 
7
 
8
  def expand2square(pil_img, background_color):
9
  width, height = pil_img.size
@@ -30,7 +31,7 @@ def process_images(
30
  new_images_aux_list = []
31
  for i, image in enumerate(images):
32
  # image.shape: (height, width, channels)
33
- print(f'@tcm: In process_images(): frame {i}')
34
  if isinstance(image, np.ndarray):
35
  image = Image.fromarray(image)
36
  image_aux_list = []
@@ -44,11 +45,9 @@ def process_images(
44
  image_aux = expand2square(
45
  image_aux, tuple(int(x * 255) for x in processor_aux.image_mean)
46
  ).resize((target_resolution, target_resolution))
47
- print(f'@tcm: In process_images(): begin processor_aux.preprocess()')
48
  image_aux = processor_aux.preprocess(image_aux, return_tensors="pt")[
49
  "pixel_values"
50
  ][0]
51
- print(f'@tcm: In process_images(): end processor_aux.preprocess()')
52
  # image_aux.shape: torch.Size([3, 384, 384])
53
  image_aux_list.append(image_aux)
54
  new_images_aux_list.append(image_aux_list) # torch.Tensor(C, H, W) new_images_aux_list[num_frames][num_processor]
@@ -83,17 +82,21 @@ def process_video_frames(
83
  image_processor: List[BaseImageProcessor],
84
  device: str
85
  ) -> Tuple[List[torch.Tensor], List[Tuple[int, int]]]:
86
- print(f'@tcm: In process_video_frames(): begin VideoReader()')
87
  vr = VideoReader(video_path, ctx=cpu(0), num_threads=1)
88
- print(f'@tcm: In process_video_frames(): end VideoReader()')
89
  fps = float(vr.get_avg_fps())
90
  frame_indices = np.array([i for i in range(0, len(vr), round(fps),)])
91
  video = []
 
92
  for frame_index in frame_indices:
93
  img = vr[frame_index].asnumpy()
94
  video.append(img)
95
  video = np.stack(video) # shape: (num_frames, height, width, channels)
 
96
  image_sizes = [video[0].shape[:2]]
 
97
  video = process_images(video, image_processor, device)
 
98
  video = [item.unsqueeze(0) for item in video]
99
  return video, image_sizes
 
4
  from decord import cpu, VideoReader
5
  from transformers import BaseImageProcessor
6
  from typing import List, Union, Tuple
7
+ import time
8
 
9
  def expand2square(pil_img, background_color):
10
  width, height = pil_img.size
 
31
  new_images_aux_list = []
32
  for i, image in enumerate(images):
33
  # image.shape: (height, width, channels)
34
+ # print(f'@tcm: In process_images(): frame {i}')
35
  if isinstance(image, np.ndarray):
36
  image = Image.fromarray(image)
37
  image_aux_list = []
 
45
  image_aux = expand2square(
46
  image_aux, tuple(int(x * 255) for x in processor_aux.image_mean)
47
  ).resize((target_resolution, target_resolution))
 
48
  image_aux = processor_aux.preprocess(image_aux, return_tensors="pt")[
49
  "pixel_values"
50
  ][0]
 
51
  # image_aux.shape: torch.Size([3, 384, 384])
52
  image_aux_list.append(image_aux)
53
  new_images_aux_list.append(image_aux_list) # torch.Tensor(C, H, W) new_images_aux_list[num_frames][num_processor]
 
82
  image_processor: List[BaseImageProcessor],
83
  device: str
84
  ) -> Tuple[List[torch.Tensor], List[Tuple[int, int]]]:
85
+ init_time = time.time()
86
  vr = VideoReader(video_path, ctx=cpu(0), num_threads=1)
87
+ print(f'@tcm: In process_video_frames(): init_time={time.time()-init_time:4f}')
88
  fps = float(vr.get_avg_fps())
89
  frame_indices = np.array([i for i in range(0, len(vr), round(fps),)])
90
  video = []
91
+ vectorize_time = time.time()
92
  for frame_index in frame_indices:
93
  img = vr[frame_index].asnumpy()
94
  video.append(img)
95
  video = np.stack(video) # shape: (num_frames, height, width, channels)
96
+ print(f'@tcm: In process_video_frames(): vectorize_time={time.time()-vectorize_time:4f}')
97
  image_sizes = [video[0].shape[:2]]
98
+ process_time = time.time()
99
  video = process_images(video, image_processor, device)
100
+ print(f'@tcm: In process_video_frames(): process_time={time.time()-process_time:4f}')
101
  video = [item.unsqueeze(0) for item in video]
102
  return video, image_sizes