今回利用したPyTorchライブラリ

PyTorch MIDAS MiDaS models for computing relative depth from a single image.

【結果】入力した画像と、出力された画像

（入力画像）

（出力画像）

ファイル保存なし
plt.show()でウィンドウ出力のみ

単眼カメラのレンズに近い近距離の物体ほど、明るい黄色で表現されている（正常動作）。

手元に画像ファイル（jpg, jpeg, png）を用意。

Terminal

electron@diynoMacBook-Pro catr % cd image_files 
electron@diynoMacBook-Pro image_files % ls
special.jp_解析結果.jpeg	test002.jpg			test005.jp_解析結果.jpeg	test008.jpe_解析結果.jpeg
special.jpg			test003.jpg			test005.jpg			test008.jpeg
test001.png			test004.jpg			test007.jpeg			test010.jpg
electron@diynoMacBook-Pro image_files %

公式サイトのサンプルコードを写経して実行

Terminal

electron@diynoMacBook-Pro image_files % pip3 install timm

Terminal

electron@diynoMacBook-Pro image_files % python3
Python 3.9.6 (default, Jun 29 2021, 06:20:32) 
[Clang 12.0.0 (clang-1200.0.32.29)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import cv2
>>> import torch
>>> import urllib.request
>>> import matplotlib.pyplot as plt
>>> 
>>> url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
>>> urllib.request.urlretrieve(url, filename)
('dog.jpg', <http.client.HTTPMessage object at 0x13352a430>)
>>> 
>>> model_type = "DPT_Large"
>>> midas = torch.hub.load("intel-isl/MiDaS", model_type)
Downloading: "https://github.com/intel-isl/MiDaS/archive/master.zip" to /Users/electron/.cache/torch/hub/master.zip
Downloading: "https://github.com/intel-isl/DPT/releases/download/1_0/dpt_large-midas-2f21e586.pt" to /Users/electron/.cache/torch/hub/checkpoints/dpt_large-midas-2f21e586.pt
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.28G/1.28G [00:31<00:00, 44.1MB/s]
>>> 
>>> device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
>>> print(device)
cpu
>>> 
>>> midas.to(device)
DPTDepthModel(
  (pretrained): Module(
    (model): VisionTransformer(
      (patch_embed): PatchEmbed(
        (proj): Conv2d(3, 1024, kernel_size=(16, 16), stride=(16, 16))
      )
      (pos_drop): Dropout(p=0.0, inplace=False)
      (blocks): ModuleList(
        (0): Block(
          (norm1): LayerNorm((1024,), eps=1e-06, elementwise_affine=True)
          (attn): Attention(
            (qkv): Linear(in_features=1024, out_features=3072, bias=True)
            (attn_drop): Dropout(p=0.0, inplace=False)
            (proj): Linear(in_features=1024, out_features=1024, bias=True)
            (proj_drop): Dropout(p=0.0, inplace=False)

（ 省略 ）

      (skip_add): FloatFunctional(
        (activation_post_process): Identity()
      )
    )
    (output_conv): Sequential(
      (0): Conv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (1): Interpolate()
      (2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (3): ReLU(inplace=True)
      (4): Conv2d(32, 1, kernel_size=(1, 1), stride=(1, 1))
      (5): ReLU(inplace=True)
      (6): Identity()
    )
  )
)
>>> 
>>> midas_transforms = torch.hub.load("intel-isl/MiDaS", "transforms")
Using cache found in /Users/electron/.cache/torch/hub/intel-isl_MiDaS_master
>>> 
>>> if model_type == "DPT_Large" or model_type == "DPT_Hybrid":
...     transform = midas_transforms.dpt_transform
... else:
...     transform = midas_transforms.small_transform
... 
>>> 
>>> print(transform)
Compose(
    <function transforms.<locals>.<lambda> at 0x13cdd5160>
    <midas.transforms.Resize object at 0x13ce46cd0>
    <midas.transforms.NormalizeImage object at 0x13ce46ca0>
    <midas.transforms.PrepareForNet object at 0x13ce46c40>
    <function transforms.<locals>.<lambda> at 0x13cdd5310>
)
>>> 
>>> img = cv2.imread("test007.jpeg")
>>> img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
>>> input_batch = transform(img).to(device)
>>> 
>>> with torch.no_grad():
...     prediction = midas(input_batch)
...     prediction = torch.nn.functional.interpolate(
...     prediction.unsqueeze(1),
...     size=img.shape[:2],
...     mode="bicubic",
...     align_corners=False,
...     ).squeeze()
... 
/usr/local/lib/python3.9/site-packages/torch/nn/functional.py:3451: UserWarning: Default upsampling behavior when mode=bilinear is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details.
  warnings.warn(

>>> 
>>> output = prediction.cpu().numpy()
>>> 
>>> import types
>>> print(type(output))
<class 'numpy.ndarray'>
>>> 
>>> plt.imshow(output)
<matplotlib.image.AxesImage object at 0x13ce6fd60>
>>> 
>>> plt.show()
>>> quit()
electron@diynoMacBook-Pro image_files %