Add another two padding methods "VALID" and "SAME" as tensorflow, and keep the existing "SAME_CLAMP_TO_EDGE" method suggested by sr filter. As "SAME_CLAMP_TO_EDGE"can keep the output with the same size as original input, and gives a slight better result as mentioned by sr filter.
Signed-off-by: Xuewei Meng <xwmen...@gmail.com> --- libavfilter/dnn_backend_native.c | 52 ++++++++++++++++++++++++-------- libavfilter/dnn_backend_native.h | 3 ++ 2 files changed, 43 insertions(+), 12 deletions(-) diff --git a/libavfilter/dnn_backend_native.c b/libavfilter/dnn_backend_native.c index 06fbdf368b..171a756385 100644 --- a/libavfilter/dnn_backend_native.c +++ b/libavfilter/dnn_backend_native.c @@ -61,6 +61,12 @@ static DNNReturnType set_input_output_native(void *model, DNNInputData *input, c return DNN_ERROR; } cur_channels = conv_params->output_num; + + if(conv_params->padding_method == VALID){ + int pad_size = conv_params->kernel_size - 1; + cur_height -= pad_size; + cur_width -= pad_size; + } break; case DEPTH_TO_SPACE: depth_to_space_params = (DepthToSpaceParams *)network->layers[layer].params; @@ -77,6 +83,10 @@ static DNNReturnType set_input_output_native(void *model, DNNInputData *input, c if (network->layers[layer].output){ av_freep(&network->layers[layer].output); } + + if(cur_height <= 0 || cur_width <= 0) + return DNN_ERROR; + network->layers[layer].output = av_malloc(cur_height * cur_width * cur_channels * sizeof(float)); if (!network->layers[layer].output){ return DNN_ERROR; @@ -154,13 +164,14 @@ DNNModel *ff_dnn_load_model_native(const char *model_filename) ff_dnn_free_model_native(&model); return NULL; } + conv_params->padding_method = (int32_t)avio_rl32(model_file_context); conv_params->activation = (int32_t)avio_rl32(model_file_context); conv_params->input_num = (int32_t)avio_rl32(model_file_context); conv_params->output_num = (int32_t)avio_rl32(model_file_context); conv_params->kernel_size = (int32_t)avio_rl32(model_file_context); kernel_size = conv_params->input_num * conv_params->output_num * conv_params->kernel_size * conv_params->kernel_size; - dnn_size += 16 + (kernel_size + conv_params->output_num << 2); + dnn_size += 20 + (kernel_size + conv_params->output_num << 2); if (dnn_size > file_size || conv_params->input_num <= 0 || conv_params->output_num <= 0 || conv_params->kernel_size <= 0){ avio_closep(&model_file_context); @@ -218,23 +229,35 @@ DNNModel *ff_dnn_load_model_native(const char *model_filename) static void convolve(const float *input, float *output, const ConvolutionalParams *conv_params, int width, int height) { - int y, x, n_filter, ch, kernel_y, kernel_x; int radius = conv_params->kernel_size >> 1; int src_linesize = width * conv_params->input_num; int filter_linesize = conv_params->kernel_size * conv_params->input_num; int filter_size = conv_params->kernel_size * filter_linesize; + int pad_size = (conv_params->padding_method == VALID) ? (conv_params->kernel_size - 1) / 2 : 0; - for (y = 0; y < height; ++y){ - for (x = 0; x < width; ++x){ - for (n_filter = 0; n_filter < conv_params->output_num; ++n_filter){ + for (int y = pad_size; y < height - pad_size; ++y){ + for (int x = pad_size; x < width - pad_size; ++x){ + for (int n_filter = 0; n_filter < conv_params->output_num; ++n_filter){ output[n_filter] = conv_params->biases[n_filter]; - for (ch = 0; ch < conv_params->input_num; ++ch){ - for (kernel_y = 0; kernel_y < conv_params->kernel_size; ++kernel_y){ - for (kernel_x = 0; kernel_x < conv_params->kernel_size; ++kernel_x){ - output[n_filter] += input[CLAMP_TO_EDGE(y + kernel_y - radius, height) * src_linesize + - CLAMP_TO_EDGE(x + kernel_x - radius, width) * conv_params->input_num + ch] * - conv_params->kernel[n_filter * filter_size + kernel_y * filter_linesize + - kernel_x * conv_params->input_num + ch]; + + for (int ch = 0; ch < conv_params->input_num; ++ch){ + for (int kernel_y = 0; kernel_y < conv_params->kernel_size; ++kernel_y){ + for (int kernel_x = 0; kernel_x < conv_params->kernel_size; ++kernel_x){ + float input_pel; + if(conv_params->padding_method == SAME_CLAMP_TO_EDGE){ + int y_pos = CLAMP_TO_EDGE(y + kernel_y - radius, height); + int x_pos = CLAMP_TO_EDGE(x + kernel_x - radius, width); + input_pel = input[y_pos * src_linesize + x_pos * conv_params->input_num + ch]; + }else{ + int y_pos = y + kernel_y - radius; + int x_pos = x + kernel_x - radius; + input_pel = (x_pos < 0 || x_pos >= width || y_pos < 0 || y_pos >= height) ? 0.0 : + input[y_pos * src_linesize + x_pos * conv_params->input_num + ch]; + } + + + output[n_filter] += input_pel * conv_params->kernel[n_filter * filter_size + kernel_y * filter_linesize + + kernel_x * conv_params->input_num + ch]; } } } @@ -305,6 +328,11 @@ DNNReturnType ff_dnn_execute_model_native(const DNNModel *model, DNNData *output conv_params = (ConvolutionalParams *)network->layers[layer].params; convolve(network->layers[layer - 1].output, network->layers[layer].output, conv_params, cur_width, cur_height); cur_channels = conv_params->output_num; + if(conv_params->padding_method == VALID){ + int pad_size = conv_params->kernel_size - 1; + cur_height -= pad_size; + cur_width -= pad_size; + } break; case DEPTH_TO_SPACE: depth_to_space_params = (DepthToSpaceParams *)network->layers[layer].params; diff --git a/libavfilter/dnn_backend_native.h b/libavfilter/dnn_backend_native.h index e13a68a168..d70cd16387 100644 --- a/libavfilter/dnn_backend_native.h +++ b/libavfilter/dnn_backend_native.h @@ -34,6 +34,8 @@ typedef enum {INPUT, CONV, DEPTH_TO_SPACE} DNNLayerType; typedef enum {RELU, TANH, SIGMOID} DNNActivationFunc; +typedef enum {VALID, SAME, SAME_CLAMP_TO_EDGE} DNNConvPaddingParam; + typedef struct Layer{ DNNLayerType type; float *output; @@ -43,6 +45,7 @@ typedef struct Layer{ typedef struct ConvolutionalParams{ int32_t input_num, output_num, kernel_size; DNNActivationFunc activation; + DNNConvPaddingParam padding_method; float *kernel; float *biases; } ConvolutionalParams; -- 2.17.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".