pt 1.9 support and hydra 1.1.0 req

README.md

@@ -46,11 +46,9 @@ When running a training inside the container it is necessary to [increase the sh
 ## Source
 1. Install CUDA (>10.1) and cudnn (make sure to select [compatible versions](https://docs.nvidia.com/deeplearning/cudnn/support-matrix/index.html)!)
 2. [Optional] Depending on your GPU you might need to set `TORCH_CUDA_ARCH_LIST`, check [compute capabilities](https://developer.nvidia.com/cuda-gpus) here.
-3. Install [torch](https://pytorch.org/) (make sure to match the pytorch and CUDA versions!) (requires pytorch >1.7+)
-4. Install [torchvision](https://github.com/pytorch/vision) (make sure to match the versions!)
-5. Clone nnDetection, `cd [path_to_repo]` and `pip install -e .`
-6. Upgrade hydra to next release: `pip install hydra-core --upgrade --pre`
-7. Set environment variables (more info can be found below):
+3. Install [torch](https://pytorch.org/) (make sure to match the pytorch and CUDA versions!) (requires pytorch >1.7+) and [torchvision](https://github.com/pytorch/vision)(make sure to match the versions!).
+4. Clone nnDetection, `cd [path_to_repo]` and `pip install -e .`
+5. Set environment variables (more info can be found below):
     - `det_data`: [required] Path to the source directory where all the data will be located
     - `det_models`: [required] Path to directory where all models will be saved
     - `OMP_NUM_THREADS=1` : [required] Needs to be set! Otherwise bad things will happen... Refer to batchgenerators documentation.

nndet/core/boxes/anchors.py

@@ -94,7 +94,7 @@ def compute_anchors_for_strides(anchors: torch.Tensor,
     return anchors_with_stride
 
 
-class AnchorGenerator2D(AnchorGenerator):
+class AnchorGenerator2D(torch.nn.Module):
     def __init__(self, sizes: Sequence[Union[int, Sequence[int]]] = (128, 256, 512),
                  aspect_ratios: Sequence[Union[float, Sequence[float]]] = (0.5, 1.0, 2.0),
                  **kwargs):
@@ -109,7 +109,18 @@ class AnchorGenerator2D(AnchorGenerator):
                 height/width, e.g. (0.5, 1, 2). if Seq[Seq] is provided, it should have
                 the same length as sizes
         """
-        super().__init__(sizes=sizes, aspect_ratios=aspect_ratios)
+        super().__init__()
+        if not isinstance(sizes[0], (list, tuple)):
+            sizes = tuple((s,) for s in sizes)
+        if not isinstance(aspect_ratios[0], (list, tuple)):
+            aspect_ratios = (aspect_ratios,) * len(sizes)
+        assert len(sizes) == len(aspect_ratios)
+
+        self.sizes = sizes
+        self.aspect_ratios = aspect_ratios
+        self.cell_anchors = None
+        self._cache = {}
+
         self.num_anchors_per_level: List[int] = None
         if kwargs:
             logger.info(f"Discarding anchor generator kwargs {kwargs}")

nndet/utils/tensor.py

@@ -5,7 +5,8 @@ import re
 import numpy as np
 from torch import Tensor
 
-from torch._six import container_abcs, string_classes, int_classes
+from collections import abc
+from torch._six import string_classes
 from typing import Sequence, Union, Any, Mapping, Callable, List
 
 np_str_obj_array_pattern = re.compile(r'[SaUO]')
@@ -129,11 +130,11 @@ def to_tensor(inp: Any) -> Any:
                 and np_str_obj_array_pattern.search(inp.dtype.str) is not None:
             return inp
         return torch.as_tensor(inp)
-    elif isinstance(inp, container_abcs.Mapping):
+    elif isinstance(inp, abc.Mapping):
         return {key: to_tensor(inp[key]) for key in inp}
     elif isinstance(inp, tuple) and hasattr(inp, '_fields'):  # namedtuple
         return elem_type(*(to_tensor(d) for d in inp))
-    elif isinstance(inp, container_abcs.Sequence) and not isinstance(inp, string_classes):
+    elif isinstance(inp, abc.Sequence) and not isinstance(inp, string_classes):
         return [to_tensor(d) for d in inp]
     else:
         return inp

requirements.txt

@@ -12,7 +12,7 @@ medpy
 SimpleITK
 tqdm
 loguru
-hydra-core
+hydra-core>=1.1.0
 mlflow
 GitPython
 matplotlib