File size: 4,087 Bytes
424188c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 |
# ------------------------------------------------------------------------------------------------
# Deformable DETR
# Copyright (c) 2020 SenseTime. All Rights Reserved.
# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
# ------------------------------------------------------------------------------------------------
# Modified from https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/tree/pytorch_1.0.0
# ------------------------------------------------------------------------------------------------
from __future__ import absolute_import
from __future__ import print_function
from __future__ import division
import time
import torch
import torch.nn as nn
from torch.autograd import gradcheck
from functions.ms_deform_attn_func import MSDeformAttnFunction, ms_deform_attn_core_pytorch
N, M, D = 1, 2, 2
Lq, L, P = 2, 2, 2
shapes = torch.as_tensor([(6, 4), (3, 2)], dtype=torch.long).cuda()
level_start_index = torch.cat((shapes.new_zeros((1, )), shapes.prod(1).cumsum(0)[:-1]))
S = sum([(H*W).item() for H, W in shapes])
torch.manual_seed(3)
@torch.no_grad()
def check_forward_equal_with_pytorch_double():
value = torch.rand(N, S, M, D).cuda() * 0.01
sampling_locations = torch.rand(N, Lq, M, L, P, 2).cuda()
attention_weights = torch.rand(N, Lq, M, L, P).cuda() + 1e-5
attention_weights /= attention_weights.sum(-1, keepdim=True).sum(-2, keepdim=True)
im2col_step = 2
output_pytorch = ms_deform_attn_core_pytorch(value.double(), shapes, sampling_locations.double(), attention_weights.double()).detach().cpu()
output_cuda = MSDeformAttnFunction.apply(value.double(), shapes, level_start_index, sampling_locations.double(), attention_weights.double(), im2col_step).detach().cpu()
fwdok = torch.allclose(output_cuda, output_pytorch)
max_abs_err = (output_cuda - output_pytorch).abs().max()
max_rel_err = ((output_cuda - output_pytorch).abs() / output_pytorch.abs()).max()
print(f'* {fwdok} check_forward_equal_with_pytorch_double: max_abs_err {max_abs_err:.2e} max_rel_err {max_rel_err:.2e}')
@torch.no_grad()
def check_forward_equal_with_pytorch_float():
value = torch.rand(N, S, M, D).cuda() * 0.01
sampling_locations = torch.rand(N, Lq, M, L, P, 2).cuda()
attention_weights = torch.rand(N, Lq, M, L, P).cuda() + 1e-5
attention_weights /= attention_weights.sum(-1, keepdim=True).sum(-2, keepdim=True)
im2col_step = 2
output_pytorch = ms_deform_attn_core_pytorch(value, shapes, sampling_locations, attention_weights).detach().cpu()
output_cuda = MSDeformAttnFunction.apply(value, shapes, level_start_index, sampling_locations, attention_weights, im2col_step).detach().cpu()
fwdok = torch.allclose(output_cuda, output_pytorch, rtol=1e-2, atol=1e-3)
max_abs_err = (output_cuda - output_pytorch).abs().max()
max_rel_err = ((output_cuda - output_pytorch).abs() / output_pytorch.abs()).max()
print(f'* {fwdok} check_forward_equal_with_pytorch_float: max_abs_err {max_abs_err:.2e} max_rel_err {max_rel_err:.2e}')
def check_gradient_numerical(channels=4, grad_value=True, grad_sampling_loc=True, grad_attn_weight=True):
value = torch.rand(N, S, M, channels).cuda() * 0.01
sampling_locations = torch.rand(N, Lq, M, L, P, 2).cuda()
attention_weights = torch.rand(N, Lq, M, L, P).cuda() + 1e-5
attention_weights /= attention_weights.sum(-1, keepdim=True).sum(-2, keepdim=True)
im2col_step = 2
func = MSDeformAttnFunction.apply
value.requires_grad = grad_value
sampling_locations.requires_grad = grad_sampling_loc
attention_weights.requires_grad = grad_attn_weight
gradok = gradcheck(func, (value.double(), shapes, level_start_index, sampling_locations.double(), attention_weights.double(), im2col_step))
print(f'* {gradok} check_gradient_numerical(D={channels})')
if __name__ == '__main__':
check_forward_equal_with_pytorch_double()
check_forward_equal_with_pytorch_float()
for channels in [30, 32, 64, 71, 1025, 2048, 3096]:
check_gradient_numerical(channels, True, True, True)
|