Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Hotfix for accuracy interfering with training #5987

Closed
wants to merge 1 commit into from
Closed

Hotfix for accuracy interfering with training #5987

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions include/caffe/layers/accuracy_layer.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,6 +92,8 @@ class AccuracyLayer : public Layer<Dtype> {
int ignore_label_;
/// Keeps counts of the number of samples per class.
Blob<Dtype> nums_buffer_;
/// Intermediate results for the GPU implementation
Blob<Dtype> gpu_buffer_;
};

} // namespace caffe
Expand Down
1 change: 1 addition & 0 deletions src/caffe/layers/accuracy_layer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@ void AccuracyLayer<Dtype>::Reshape(
<< "label count (number of labels) must be N*H*W, "
<< "with integer values in {0, 1, ..., C-1}.";
vector<int> top_shape(0); // Accuracy is a scalar; 0 axes.
gpu_buffer_.ReshapeLike(*bottom[0]);
top[0]->Reshape(top_shape);
if (top.size() > 1) {
// Per-class accuracy is a vector; 1 axes.
Expand Down
11 changes: 3 additions & 8 deletions src/caffe/layers/accuracy_layer.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,16 +71,11 @@ void AccuracyLayer<Dtype>::Forward_gpu(
const int dim = bottom[0]->count() / outer_num_;
const int num_labels = bottom[0]->shape(label_axis_);
const int nthreads = outer_num_ * inner_num_;
// Since this memory is not used for anything,
// we use it here to avoid having to allocate new GPU
// memory to accumulate intermediate results in the kernel.
Dtype* acc_data = bottom[0]->mutable_gpu_diff();
Dtype* acc_data = gpu_buffer_.mutable_gpu_data();
if (top.size() == 1) {
// simple case - report only global accuracy.

// Similarly, this memory is never used elsewhere, and thus we can use it
// to avoid having to allocate additional GPU memory.
Dtype* counts = bottom[1]->mutable_gpu_diff();
Dtype* counts = gpu_buffer_.mutable_gpu_diff();
// NOLINT_NEXT_LINE(whitespace/operators)
AccuracyForwardGPU<Dtype><<<CAFFE_GET_BLOCKS(nthreads),
CAFFE_CUDA_NUM_THREADS>>>(nthreads, bottom_data, bottom_label,
Expand Down Expand Up @@ -113,7 +108,7 @@ void AccuracyLayer<Dtype>::Forward_gpu(

// get the overall accuracy
Dtype acc;
caffe_gpu_asum(bottom[0]->count(), acc_data, &acc);
caffe_gpu_asum(gpu_buffer_.count(), acc_data, &acc);
Dtype valid_count;
caffe_gpu_asum(nums_buffer_.count(), counts, &valid_count);
if (valid_count > 0) {
Expand Down