In <dl_matrix3dff_conv_common> function the bias weights are not taken into account.

espressif / esp-dl

Espressif deep-learning library for AIoT applications

MIT License

548 stars 118 forks source link

In <dl_matrix3dff_conv_common> function the bias weights are not taken into account. #26

Closed Vvard closed 4 years ago

Vvard commented 4 years ago

input cnn3x3_with_1_bias_weight cnn3x3_with_0_bias_weight one conv layer conv_output_with_1_bias_weight conv_output_with_0_bias_weight

I used "dl_matrix3dff_conv_common(dl_matrix3d_t in, dl_matrix3d_t filter, dl_matrix3d_t *bias, int stride_x, int stride_y, dl_padding_type padding)" function with different bias weights, but there are no changing results. This is a small test program with 4x4 input data, where I run the program in the case of 0 bias weight and in the case of 1 bias weight and I get the same result.

XiaochaoGONG commented 4 years ago

Good test ! I'll fix that.

XiaochaoGONG commented 4 years ago

f1a3dbe8ffe418aa6f9372831f9c1e33a651ab5b Please update and try.

Vvard commented 4 years ago

Works correctly! thanks)