From patchwork Wed Aug 14 07:06:00 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Guo, Yejun" X-Patchwork-Id: 14503 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id 6FADC449B8D for ; Wed, 14 Aug 2019 10:09:43 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 5654168AA75; Wed, 14 Aug 2019 10:09:43 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mga12.intel.com (mga12.intel.com [192.55.52.136]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 571E868A5EE for ; Wed, 14 Aug 2019 10:09:40 +0300 (EEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga003.jf.intel.com ([10.7.209.27]) by fmsmga106.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 14 Aug 2019 00:09:39 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.64,384,1559545200"; d="scan'208";a="178923024" Received: from yguo18-skl-u1604.sh.intel.com ([10.239.13.25]) by orsmga003.jf.intel.com with ESMTP; 14 Aug 2019 00:09:37 -0700 From: "Guo, Yejun" To: ffmpeg-devel@ffmpeg.org Date: Wed, 14 Aug 2019 15:06:00 +0800 Message-Id: <1565766360-31048-1-git-send-email-yejun.guo@intel.com> X-Mailer: git-send-email 2.7.4 Subject: [FFmpeg-devel] [PATCH 2/2] libavfilter/dnn/dnn_backend_tf: add tf.pad support for tensorflow backend with native model. X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: yejun.guo@intel.com MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Signed-off-by: Guo, Yejun --- libavfilter/dnn/dnn_backend_tf.c | 47 ++++++++++++++++------------------------ 1 file changed, 19 insertions(+), 28 deletions(-) diff --git a/libavfilter/dnn/dnn_backend_tf.c b/libavfilter/dnn/dnn_backend_tf.c index ca7434a..75dbe5e 100644 --- a/libavfilter/dnn/dnn_backend_tf.c +++ b/libavfilter/dnn/dnn_backend_tf.c @@ -27,6 +27,7 @@ #include "dnn_backend_native.h" #include "libavformat/avio.h" #include "libavutil/avassert.h" +#include "dnn_backend_native_layer_pad.h" #include @@ -347,23 +348,8 @@ static DNNReturnType add_depth_to_space_layer(TFModel *tf_model, TF_Operation ** return DNN_SUCCESS; } -static int calculate_pad(const ConvolutionalNetwork *conv_network) -{ - ConvolutionalParams *params; - int32_t layer; - int pad = 0; - - for (layer = 0; layer < conv_network->layers_num; ++layer){ - if (conv_network->layers[layer].type == CONV){ - params = (ConvolutionalParams *)conv_network->layers[layer].params; - pad += params->kernel_size >> 1; - } - } - - return pad; -} - -static DNNReturnType add_pad_op(TFModel *tf_model, TF_Operation **cur_op, const int32_t pad) +static DNNReturnType add_pad_layer(TFModel *tf_model, TF_Operation **cur_op, + LayerPadParams *params, const int layer) { TF_Operation *op; TF_Tensor *tensor; @@ -374,14 +360,21 @@ static DNNReturnType add_pad_op(TFModel *tf_model, TF_Operation **cur_op, const input.index = 0; - op_desc = TF_NewOperation(tf_model->graph, "Const", "pads"); + char name_buffer[NAME_BUFFER_SIZE]; + snprintf(name_buffer, NAME_BUFFER_SIZE, "pad%d", layer); + + op_desc = TF_NewOperation(tf_model->graph, "Const", name_buffer); TF_SetAttrType(op_desc, "dtype", TF_INT32); tensor = TF_AllocateTensor(TF_INT32, pads_shape, 2, 4 * 2 * sizeof(int32_t)); pads = (int32_t *)TF_TensorData(tensor); - pads[0] = 0; pads[1] = 0; - pads[2] = pad; pads[3] = pad; - pads[4] = pad; pads[5] = pad; - pads[6] = 0; pads[7] = 0; + pads[0] = params->paddings[0][0]; + pads[1] = params->paddings[0][1]; + pads[2] = params->paddings[1][0]; + pads[3] = params->paddings[1][1]; + pads[4] = params->paddings[2][0]; + pads[5] = params->paddings[2][1]; + pads[6] = params->paddings[3][0]; + pads[7] = params->paddings[3][1]; TF_SetAttrTensor(op_desc, "value", tensor, tf_model->status); if (TF_GetCode(tf_model->status) != TF_OK){ return DNN_ERROR; @@ -418,7 +411,6 @@ static DNNReturnType load_native_model(TFModel *tf_model, const char *model_file int32_t *transpose_perm; int64_t transpose_perm_shape[] = {4}; int64_t input_shape[] = {1, -1, -1, -1}; - int32_t pad; DNNReturnType layer_add_res; DNNModel *native_model = NULL; ConvolutionalNetwork *conv_network; @@ -429,7 +421,6 @@ static DNNReturnType load_native_model(TFModel *tf_model, const char *model_file } conv_network = (ConvolutionalNetwork *)native_model->model; - pad = calculate_pad(conv_network); tf_model->graph = TF_NewGraph(); tf_model->status = TF_NewStatus(); @@ -448,10 +439,6 @@ static DNNReturnType load_native_model(TFModel *tf_model, const char *model_file CLEANUP_ON_ERROR(tf_model); } - if (add_pad_op(tf_model, &op, pad) != DNN_SUCCESS){ - CLEANUP_ON_ERROR(tf_model); - } - op_desc = TF_NewOperation(tf_model->graph, "Const", "transpose_perm"); TF_SetAttrType(op_desc, "dtype", TF_INT32); tensor = TF_AllocateTensor(TF_INT32, transpose_perm_shape, 1, 4 * sizeof(int32_t)); @@ -479,6 +466,10 @@ static DNNReturnType load_native_model(TFModel *tf_model, const char *model_file layer_add_res = add_depth_to_space_layer(tf_model, &op, (DepthToSpaceParams *)conv_network->layers[layer].params, layer); break; + case MIRROR_PAD: + layer_add_res = add_pad_layer(tf_model, &op, + (LayerPadParams *)conv_network->layers[layer].params, layer); + break; default: CLEANUP_ON_ERROR(tf_model); }