Not implemented on the CPU

[Shamaseen-iotistic]

I have tried to run the code on the CPU by adding this line of code:

cfg = setup_cfg(config_file)
cfg.defrost()
cfg.MODEL.DEVICE='cpu'

but i got this error :

DeepSolo/adet/layers/ms_deform_attn.py:24, in _MSDeformAttnFunction.forward(ctx, value, value_spatial_shapes, value_level_start_index, sampling_locations, attention_weights, im2col_step)
     21 @staticmethod
     22 def forward(ctx, value, value_spatial_shapes, value_level_start_index, sampling_locations, attention_weights, im2col_step):
     23     ctx.im2col_step = im2col_step
---> 24     output = _C.ms_deform_attn_forward(
     25         value, value_spatial_shapes, value_level_start_index, sampling_locations, attention_weights, ctx.im2col_step)
     26     ctx.save_for_backward(value, value_spatial_shapes, value_level_start_index, sampling_locations, attention_weights)
     27     return output
RuntimeError: Not implemented on the CPU

is there any way to run this code on the CPU???

[EvilGarfield]

Facing the same issue. A workaround for using this on CPU only would be greatly appreciated.

[EvilGarfield]

I solved the issue by adding a definition for the multi scale deformable in .\adet\layers\ms_deform_attn.py

def multi_scale_deformable_attn_pytorch(
        value: torch.Tensor, value_spatial_shapes: torch.Tensor,
        sampling_locations: torch.Tensor,
        attention_weights: torch.Tensor) -> torch.Tensor:
    """CPU version of multi-scale deformable attention.

    Args:
        value (torch.Tensor): The value has shape
            (bs, num_keys, num_heads, embed_dims//num_heads)
        value_spatial_shapes (torch.Tensor): Spatial shape of
            each feature map, has shape (num_levels, 2),
            last dimension 2 represent (h, w)
        sampling_locations (torch.Tensor): The location of sampling points,
            has shape
            (bs ,num_queries, num_heads, num_levels, num_points, 2),
            the last dimension 2 represent (x, y).
        attention_weights (torch.Tensor): The weight of sampling points used
            when calculate the attention, has shape
            (bs ,num_queries, num_heads, num_levels, num_points),

    Returns:
        torch.Tensor: has shape (bs, num_queries, embed_dims)
    """

    bs, _, num_heads, embed_dims = value.shape
    _, num_queries, num_heads, num_levels, num_points, _ =\
        sampling_locations.shape
    value_list = value.split([H_ * W_ for H_, W_ in value_spatial_shapes],
                             dim=1)
    sampling_grids = 2 * sampling_locations - 1
    sampling_value_list = []
    for level, (H_, W_) in enumerate(value_spatial_shapes):
        # bs, H_*W_, num_heads, embed_dims ->
        # bs, H_*W_, num_heads*embed_dims ->
        # bs, num_heads*embed_dims, H_*W_ ->
        # bs*num_heads, embed_dims, H_, W_
        value_l_ = value_list[level].flatten(2).transpose(1, 2).reshape(
            bs * num_heads, embed_dims, H_, W_)
        # bs, num_queries, num_heads, num_points, 2 ->
        # bs, num_heads, num_queries, num_points, 2 ->
        # bs*num_heads, num_queries, num_points, 2
        sampling_grid_l_ = sampling_grids[:, :, :,
                                          level].transpose(1, 2).flatten(0, 1)
        # bs*num_heads, embed_dims, num_queries, num_points
        sampling_value_l_ = F.grid_sample(
            value_l_,
            sampling_grid_l_,
            mode='bilinear',
            padding_mode='zeros',
            align_corners=False)
        sampling_value_list.append(sampling_value_l_)
    # (bs, num_queries, num_heads, num_levels, num_points) ->
    # (bs, num_heads, num_queries, num_levels, num_points) ->
    # (bs, num_heads, 1, num_queries, num_levels*num_points)
    attention_weights = attention_weights.transpose(1, 2).reshape(
        bs * num_heads, 1, num_queries, num_levels * num_points)
    output = (torch.stack(sampling_value_list, dim=-2).flatten(-2) *
              attention_weights).sum(-1).view(bs, num_heads * embed_dims,
                                              num_queries)
    return output.transpose(1, 2).contiguous()

and then changed the line

class _MSDeformAttnFunction(torch.autograd.Function):
    @staticmethod
    ...
        output = _C.ms_deform_attn_forward(
            value, value_spatial_shapes, value_level_start_index, sampling_locations, attention_weights, ctx.im2col_step)
    ...

with
output = multi_scale_deformable_attn_pytorch(value, value_spatial_shapes, sampling_locations, attention_weights)