WZMIAOMIAO/deep-learning-for-image-processing · error · ValueError

images is expected to be a list of 3d tensors of shape [C, H

Error message

images is expected to be a list of 3d tensors of shape [C, H, W], got {}

What it means

GeneralizedRCNNTransform.forward expects each input image to be a 3-D CHW tensor (channels, height, width). If any image has dim != 3 — usually because a batched 4-D tensor or a 2-D grayscale tensor was passed — it raises ValueError with the offending shape. Normalization and resizing operate per-image, so batching must happen outside the transform.

Source

Thrown at pytorch_object_detection/retinaNet/network_files/transform.py:243

        _indent = '\n    '
        format_string += "{0}Normalize(mean={1}, std={2})".format(_indent, self.image_mean, self.image_std)
        format_string += "{0}Resize(min_size={1}, max_size={2}, mode='bilinear')".format(_indent, self.min_size,
                                                                                         self.max_size)
        format_string += '\n)'
        return format_string

    def forward(self,
                images,       # type: List[Tensor]
                targets=None  # type: Optional[List[Dict[str, Tensor]]]
                ):
        # type: (...) -> Tuple[ImageList, Optional[List[Dict[str, Tensor]]]]
        images = [img for img in images]
        for i in range(len(images)):
            image = images[i]
            target_index = targets[i] if targets is not None else None

            if image.dim() != 3:
                raise ValueError("images is expected to be a list of 3d tensors "
                                 "of shape [C, H, W], got {}".format(image.shape))
            image = self.normalize(image)                # 对图像进行标准化处理
            image, target_index = self.resize(image, target_index)   # 对图像和对应的bboxes缩放到指定范围
            images[i] = image
            if targets is not None and target_index is not None:
                targets[i] = target_index

        # 记录resize后的图像尺寸
        image_sizes = [img.shape[-2:] for img in images]
        images = self.batch_images(images)  # 将images打包成一个batch
        image_sizes_list = torch.jit.annotate(List[Tuple[int, int]], [])

        for image_size in image_sizes:
            assert len(image_size) == 2
            image_sizes_list.append((image_size[0], image_size[1]))

        image_list = ImageList(images, image_sizes_list)
        return image_list, targets

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Pass a list of individual [C, H, W] tensors: model([img1, img2], targets) rather than a stacked batch.
2. Convert grayscale to RGB: Image.open(p).convert('RGB') so tensors are always 3-D.
3. Convert HWC to CHW: img.permute(2, 0, 1) or use transforms.ToTensor() which does it.
4. If you have a batched tensor, split it: [images[i] for i in range(images.shape[0])].

Example fix

// before
imgs, targets = next(iterator)        # imgs: [B, C, H, W] stacked by collate
model(imgs, targets)
// after
images = [img for img in imgs]        # list of [C, H, W]
model(images, targets)

assert all(isinstance(img, torch.Tensor) and img.dim() == 3 for img in images), "each image must be a [C, H, W] tensor"

def is_chw_image(img) -> bool:
    import torch
    return isinstance(img, torch.Tensor) and img.dim() == 3 and img.shape[0] in (1, 3)

try:
    outputs = model(images, targets)
except ValueError as e:
    if "3d tensors of shape [C, H, W]" in str(e):
        images = [im.squeeze(0) if im.dim() == 4 else im.permute(2, 0, 1) if im.dim() == 3 else im for im in [batched_or_raw]]
        outputs = model(images, targets)
    else:
        raise

When it happens

Trigger: Passing a DataLoader-collated [B, C, H, W] batch instead of a list of 3-D tensors; loading grayscale images with PIL convert('L') yielding [H, W]; passing the raw output of cv2.imread ([H, W, C], HWC instead of CHW).

Common situations: Custom training loops that batch with default collate; forgetting ToTensor() or image_layout conversion; grayscale datasets; mixing image-list API (torchvision detection models) with the standard batched API of classification models.

Understand the failure class

Background: Tensor shape mismatch errors ("must have shape", "expected shape ... got ..."): when tensor dimensions disagree with what an op or layer was told to expect — this error's family across 6 libraries.

Related errors

• Expected target boxes to be a tensorof shape [N, 4], got {:}
• Expected target boxes to be a tensorof shape [N, 4], got {:}
• No ground-truth boxes available for one of the images during
• No proposal boxes available for one of the images during tra
• backbone should contain an attribute out_channels specifying

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/7938aa80f53af1fb. Report an issue: GitHub.