From patchwork Mon Feb 22 07:30:45 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Guo, Yejun" X-Patchwork-Id: 25886 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 C5BC7448857 for ; Mon, 22 Feb 2021 09:41:29 +0200 (EET) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id A8FA268A8F3; Mon, 22 Feb 2021 09:41:29 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mga05.intel.com (mga05.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 7618E68A199 for ; Mon, 22 Feb 2021 09:41:22 +0200 (EET) IronPort-SDR: geTC4Idg9V8Hcmg7i5JyX4faj9H3aHhxpqAjii6JNsNsO0sToyPouwsi+L4B1d6MPO207mouB4 5WDrYvabdbJA== X-IronPort-AV: E=McAfee;i="6000,8403,9902"; a="269295460" X-IronPort-AV: E=Sophos;i="5.81,196,1610438400"; d="scan'208";a="269295460" Received: from orsmga001.jf.intel.com ([10.7.209.18]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 21 Feb 2021 23:41:21 -0800 IronPort-SDR: D9pFPtpCbKuCUYVlkQ1IM4lgx/K5c26klz68h4I0sy4F082Wr8AFYcDBQe/sx211NFvJV3AFsa u6Xj3u5UJpUg== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.81,196,1610438400"; d="scan'208";a="441301010" Received: from yguo18-skl-u1604.sh.intel.com ([10.239.159.53]) by orsmga001.jf.intel.com with ESMTP; 21 Feb 2021 23:41:18 -0800 From: "Guo, Yejun" To: ffmpeg-devel@ffmpeg.org Date: Mon, 22 Feb 2021 15:30:45 +0800 Message-Id: <20210222073045.16070-3-yejun.guo@intel.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: <20210222073045.16070-1-yejun.guo@intel.com> References: <20210222073045.16070-1-yejun.guo@intel.com> Subject: [FFmpeg-devel] [PATCH V3 3/3] libavfilter: add filter dnn_detect for object detection 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" Below are the example steps to do object detection: 1. download and install l_openvino_toolkit_p_2021.1.110.tgz from https://software.intel.com/content/www/us/en/develop/tools/openvino-toolkit/download.html or, we can get source code (tag 2021.1), build and install. 2. export LD_LIBRARY_PATH with openvino settings, for example: .../deployment_tools/inference_engine/lib/intel64/:.../deployment_tools/inference_engine/external/tbb/lib/ 3. rebuild ffmpeg from source code with configure option: --enable-libopenvino --extra-cflags='-I.../deployment_tools/inference_engine/include/' --extra-ldflags='-L.../deployment_tools/inference_engine/lib/intel64' 4. download model files and test image wget https://github.com/guoyejun/ffmpeg_dnn/raw/main/models/openvino/2021.1/face-detection-adas-0001.bin wget https://github.com/guoyejun/ffmpeg_dnn/raw/main/models/openvino/2021.1/face-detection-adas-0001.xml wget https://github.com/guoyejun/ffmpeg_dnn/raw/main/models/openvino/2021.1/face-detection-adas-0001.label wget https://github.com/guoyejun/ffmpeg_dnn/raw/main/images/cici.jpg 5. run ffmpeg with: ./ffmpeg -i cici.jpg -vf dnn_detect=dnn_backend=openvino:model=face-detection-adas-0001.xml:input=data:output=detection_out:confidence=0.6:labels=face-detection-adas-0001.label -f null - We'll see the detect result as below: [Parsed_dnn_detect_0 @ 0x56226644e540] frame->private_ref (bounding boxes): [Parsed_dnn_detect_0 @ 0x56226644e540] source: face-detection-adas-0001.xml [Parsed_dnn_detect_0 @ 0x56226644e540] index: 0, region: (1005, 813) -> (1086, 905), label: face, confidence: 10000/10000. [Parsed_dnn_detect_0 @ 0x56226644e540] index: 1, region: (888, 839) -> (967, 926), label: face, confidence: 6917/10000. There are two faces detected with confidence 100% and 69.17%. Signed-off-by: Guo, Yejun --- configure | 1 + doc/filters.texi | 40 +++ libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/dnn/dnn_backend_openvino.c | 12 + libavfilter/dnn_filter_common.c | 7 + libavfilter/dnn_filter_common.h | 1 + libavfilter/dnn_interface.h | 6 +- libavfilter/vf_dnn_detect.c | 462 +++++++++++++++++++++++++ 9 files changed, 529 insertions(+), 2 deletions(-) create mode 100644 libavfilter/vf_dnn_detect.c diff --git a/configure b/configure index 336301cb40..cdac292c2f 100755 --- a/configure +++ b/configure @@ -3549,6 +3549,7 @@ derain_filter_select="dnn" deshake_filter_select="pixelutils" deshake_opencl_filter_deps="opencl" dilation_opencl_filter_deps="opencl" +dnn_detect_filter_select="dnn" dnn_processing_filter_select="dnn" drawtext_filter_deps="libfreetype" drawtext_filter_suggest="libfontconfig libfribidi" diff --git a/doc/filters.texi b/doc/filters.texi index 426cb158da..55368c6f1b 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -10133,6 +10133,46 @@ ffmpeg -i INPUT -f lavfi -i nullsrc=hd720,geq='r=128+80*(sin(sqrt((X-W/2)*(X-W/2 @end example @end itemize +@section dnn_detect + +Do object detection with deep neural networks. + +The filter accepts the following options: + +@table @option +@item dnn_backend +Specify which DNN backend to use for model loading and execution. This option accepts +only openvino now, tensorflow backends will be added. + +@item model +Set path to model file specifying network architecture and its parameters. +Note that different backends use different file formats. + +@item input +Set the input name of the dnn network. + +@item output +Set the output name of the dnn network. + +@item confidence +Set the confidence threshold (default: 0.5). + +@item labels +Set path to label file specifying the mapping between label id and name. +Each label name is written in one line, tailing spaces and empty lines are skipped. +The first line is the name of label id 0 (usually it is 'background'), +and the second line is the name of label id 1, etc. +The label id is considered as name if the label file is not provided. + +@item backend_configs +Set the configs to be passed into backend + +@item async +use DNN async execution if set (default: set), +roll back to sync execution if the backend does not support async. + +@end table + @anchor{dnn_processing} @section dnn_processing diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 359ea7f903..b14c0ecdb9 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -245,6 +245,7 @@ OBJS-$(CONFIG_DILATION_FILTER) += vf_neighbor.o OBJS-$(CONFIG_DILATION_OPENCL_FILTER) += vf_neighbor_opencl.o opencl.o \ opencl/neighbor.o OBJS-$(CONFIG_DISPLACE_FILTER) += vf_displace.o framesync.o +OBJS-$(CONFIG_DNN_DETECT_FILTER) += vf_dnn_detect.o OBJS-$(CONFIG_DNN_PROCESSING_FILTER) += vf_dnn_processing.o OBJS-$(CONFIG_DOUBLEWEAVE_FILTER) += vf_weave.o OBJS-$(CONFIG_DRAWBOX_FILTER) += vf_drawbox.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 452c030706..85419ec3f1 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -230,6 +230,7 @@ extern AVFilter ff_vf_detelecine; extern AVFilter ff_vf_dilation; extern AVFilter ff_vf_dilation_opencl; extern AVFilter ff_vf_displace; +extern AVFilter ff_vf_dnn_detect; extern AVFilter ff_vf_dnn_processing; extern AVFilter ff_vf_doubleweave; extern AVFilter ff_vf_drawbox; diff --git a/libavfilter/dnn/dnn_backend_openvino.c b/libavfilter/dnn/dnn_backend_openvino.c index 5be053b7f8..928d84b744 100644 --- a/libavfilter/dnn/dnn_backend_openvino.c +++ b/libavfilter/dnn/dnn_backend_openvino.c @@ -621,6 +621,12 @@ DNNReturnType ff_dnn_execute_model_ov(const DNNModel *model, const char *input_n return DNN_ERROR; } + if (model->func_type != DFT_PROCESS_FRAME) { + if (!out_frame) { + out_frame = in_frame; + } + } + if (nb_output != 1) { // currently, the filter does not need multiple outputs, // so we just pending the support until we really need it. @@ -674,6 +680,12 @@ DNNReturnType ff_dnn_execute_model_async_ov(const DNNModel *model, const char *i return DNN_ERROR; } + if (model->func_type != DFT_PROCESS_FRAME) { + if (!out_frame) { + out_frame = in_frame; + } + } + task = av_malloc(sizeof(*task)); if (!task) { av_log(ctx, AV_LOG_ERROR, "unable to alloc memory for task item.\n"); diff --git a/libavfilter/dnn_filter_common.c b/libavfilter/dnn_filter_common.c index 413adba406..92b696e710 100644 --- a/libavfilter/dnn_filter_common.c +++ b/libavfilter/dnn_filter_common.c @@ -64,6 +64,13 @@ int ff_dnn_init(DnnContext *ctx, DNNFunctionType func_type, AVFilterContext *fil return 0; } +int ff_dnn_set_proc(DnnContext *ctx, PRE_POST_PROC pre_proc, PRE_POST_PROC post_proc) +{ + ctx->model->pre_proc = pre_proc; + ctx->model->post_proc = post_proc; + return 0; +} + DNNReturnType ff_dnn_get_input(DnnContext *ctx, DNNData *input) { return ctx->model->get_input(ctx->model->model, input, ctx->model_inputname); diff --git a/libavfilter/dnn_filter_common.h b/libavfilter/dnn_filter_common.h index 79c4d3efe3..0e88b88bdd 100644 --- a/libavfilter/dnn_filter_common.h +++ b/libavfilter/dnn_filter_common.h @@ -48,6 +48,7 @@ typedef struct DnnContext { int ff_dnn_init(DnnContext *ctx, DNNFunctionType func_type, AVFilterContext *filter_ctx); +int ff_dnn_set_proc(DnnContext *ctx, PRE_POST_PROC pre_proc, PRE_POST_PROC post_proc); DNNReturnType ff_dnn_get_input(DnnContext *ctx, DNNData *input); DNNReturnType ff_dnn_get_output(DnnContext *ctx, int input_width, int input_height, int *output_width, int *output_height); DNNReturnType ff_dnn_execute_model(DnnContext *ctx, AVFrame *in_frame, AVFrame *out_frame); diff --git a/libavfilter/dnn_interface.h b/libavfilter/dnn_interface.h index d3a0c58a61..90a08129f4 100644 --- a/libavfilter/dnn_interface.h +++ b/libavfilter/dnn_interface.h @@ -63,6 +63,8 @@ typedef struct DNNData{ DNNColorOrder order; } DNNData; +typedef int (*PRE_POST_PROC)(AVFrame *frame, DNNData *model, AVFilterContext *filter_ctx); + typedef struct DNNModel{ // Stores model that can be different for different backends. void *model; @@ -80,10 +82,10 @@ typedef struct DNNModel{ const char *output_name, int *output_width, int *output_height); // set the pre process to transfer data from AVFrame to DNNData // the default implementation within DNN is used if it is not provided by the filter - int (*pre_proc)(AVFrame *frame_in, DNNData *model_input, AVFilterContext *filter_ctx); + PRE_POST_PROC pre_proc; // set the post process to transfer data from DNNData to AVFrame // the default implementation within DNN is used if it is not provided by the filter - int (*post_proc)(AVFrame *frame_out, DNNData *model_output, AVFilterContext *filter_ctx); + PRE_POST_PROC post_proc; } DNNModel; // Stores pointers to functions for loading, executing, freeing DNN models for one of the backends. diff --git a/libavfilter/vf_dnn_detect.c b/libavfilter/vf_dnn_detect.c new file mode 100644 index 0000000000..13b82579a9 --- /dev/null +++ b/libavfilter/vf_dnn_detect.c @@ -0,0 +1,462 @@ +/* + * Copyright (c) 2021 + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file + * implementing an object detecting filter using deep learning networks. + */ + +#include "libavformat/avio.h" +#include "libavutil/opt.h" +#include "libavutil/pixdesc.h" +#include "libavutil/avassert.h" +#include "libavutil/imgutils.h" +#include "filters.h" +#include "dnn_filter_common.h" +#include "formats.h" +#include "internal.h" +#include "libavutil/time.h" +#include "bbox.h" + +typedef struct DnnDetectContext { + const AVClass *class; + DnnContext dnnctx; + float confidence; + char *labels_filename; + char **labels; + int label_count; +} DnnDetectContext; + +#define OFFSET(x) offsetof(DnnDetectContext, dnnctx.x) +#define OFFSET2(x) offsetof(DnnDetectContext, x) +#define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM +static const AVOption dnn_detect_options[] = { + { "dnn_backend", "DNN backend", OFFSET(backend_type), AV_OPT_TYPE_INT, { .i64 = 2 }, INT_MIN, INT_MAX, FLAGS, "backend" }, +#if (CONFIG_LIBOPENVINO == 1) + { "openvino", "openvino backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, 0, 0, FLAGS, "backend" }, +#endif + DNN_COMMON_OPTIONS + { "confidence", "threshold of confidence", OFFSET2(confidence), AV_OPT_TYPE_FLOAT, { .dbl = 0.5 }, 0, 1, FLAGS}, + { "labels", "path to labels file", OFFSET2(labels_filename), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS }, + { NULL } +}; + +AVFILTER_DEFINE_CLASS(dnn_detect); + +// remove this function once we update vf_drawbox/text +// to visualize the bbox +static void dump_boundingbox(AVFilterContext *ctx, AVFrame *frame) +{ + int nb_bbox = 0; + BoundingBox *bbox; + BoundingBoxHeader *header; + + if (!frame->private_ref) + return; + + header = (BoundingBoxHeader *)frame->private_ref->data; + bbox = (BoundingBox *)(header + 1); + + if (!header->bbox_size || (frame->private_ref->size - sizeof(*header)) % header->bbox_size != 0) { + av_log(ctx, AV_LOG_ERROR, "invalid size of bounding box\n"); + return; + } + + nb_bbox = (frame->private_ref->size - sizeof(*header)) / header->bbox_size; + av_log(ctx, AV_LOG_INFO, "frame->private_ref (bounding boxes):\n"); + av_log(ctx, AV_LOG_INFO, "source: %s\n", header->source); + for (int i = 0; i < nb_bbox; i++) { + av_log(ctx, AV_LOG_INFO, "index: %d, region: (%d, %d) -> (%d, %d), label: %s, confidence: %d/%d.\n", + i, bbox->left, bbox->top, bbox->right, bbox->bottom, + bbox->detect_label, bbox->detect_confidence.num, bbox->detect_confidence.den); + if (bbox->classify_count > 0) { + for (int j = 0; j < bbox->classify_count; j++) { + av_log(ctx, AV_LOG_INFO, "\t\tclassify: label: %s, confidence: %d/%d.\n", + bbox->classify_labels[j], bbox->classify_confidences[j].num, bbox->classify_confidences[j].den); + } + } + bbox++; + } +} + +static int dnn_detect_post_proc(AVFrame *frame, DNNData *output, AVFilterContext *filter_ctx) +{ + DnnDetectContext *ctx = filter_ctx->priv; + float conf_threshold = ctx->confidence; + int proposal_count = output->height; + int detect_size = output->width; + float *detections = output->data; + int nb_bbox = 0; + BoundingBox *bbox; + BoundingBoxHeader *header; + + for (int i = 0; i < proposal_count; ++i) { + float conf = detections[i * detect_size + 2]; + if (conf < conf_threshold) { + continue; + } + nb_bbox++; + } + + if (nb_bbox == 0) { + av_log(filter_ctx, AV_LOG_VERBOSE, "nothing detected in this frame.\n"); + return 0; + } + + if (frame->private_ref) { + av_log(filter_ctx, AV_LOG_ERROR, "frame->private_ref is already occupied\n"); + return -1; + } + + frame->private_ref = av_buffer_alloc(sizeof(*header) + sizeof(*bbox) * nb_bbox); + if (!frame->private_ref) { + av_log(filter_ctx, AV_LOG_ERROR, "failed to allocate buffer for %d bboxes\n", nb_bbox); + return AVERROR(ENOMEM);; + } + + header = (BoundingBoxHeader *)frame->private_ref->data; + strncpy(header->source, ctx->dnnctx.model_filename, sizeof(header->source)); + header->source[sizeof(header->source) - 1] = '\0'; + header->bbox_size = sizeof(*bbox); + + bbox = (BoundingBox *)(header + 1); + for (int i = 0; i < proposal_count; ++i) { + int av_unused image_id = (int)detections[i * detect_size + 0]; + int label_id = (int)detections[i * detect_size + 1]; + float conf = detections[i * detect_size + 2]; + float x0 = detections[i * detect_size + 3]; + float y0 = detections[i * detect_size + 4]; + float x1 = detections[i * detect_size + 5]; + float y1 = detections[i * detect_size + 6]; + + if (conf < conf_threshold) { + continue; + } + + bbox->left = (int)(x0 * frame->width); + bbox->right = (int)(x1 * frame->width); + bbox->top = (int)(y0 * frame->height); + bbox->bottom = (int)(y1 * frame->height); + + bbox->detect_confidence = av_make_q((int)(conf * 10000), 10000); + bbox->classify_count = 0; + + if (ctx->labels && label_id < ctx->label_count) { + strcpy(bbox->detect_label, ctx->labels[label_id]); + } else { + snprintf(bbox->detect_label, BBOX_LABEL_NAME_MAX_LENGTH, "%d", label_id); + } + + nb_bbox--; + if (nb_bbox == 0) { + break; + } + bbox++; + } + + dump_boundingbox(filter_ctx, frame); + + return 0; +} + +static void free_detect_labels(DnnDetectContext *ctx) +{ + for (int i = 0; i < ctx->label_count; i++) { + av_freep(&ctx->labels[i]); + } + av_freep(&ctx->labels); +} + +static int read_detect_label_file(AVFilterContext *context) +{ + int line_len; + FILE *file; + DnnDetectContext *ctx = context->priv; + + file = av_fopen_utf8(ctx->labels_filename, "r"); + if (!file){ + av_log(context, AV_LOG_ERROR, "failed to open file %s\n", ctx->labels_filename); + return AVERROR(EINVAL); + } + + while (!feof(file)) { + char *label; + char buf[256]; + if (!fgets(buf, 256, file)) { + break; + } + + line_len = strlen(buf); + while (line_len) { + int i = line_len - 1; + if (buf[i] == '\n' || buf[i] == '\r' || buf[i] == ' ') { + buf[i] = '\0'; + line_len--; + } else { + break; + } + } + + if (line_len == 0) // empty line + continue; + + if (line_len > BBOX_LABEL_NAME_MAX_LENGTH) { + av_log(context, AV_LOG_ERROR, "label %s too long\n", buf); + fclose(file); + free_detect_labels(ctx); + return AVERROR(EINVAL); + } + + label = av_strdup(buf); + if (!label) { + av_log(context, AV_LOG_ERROR, "failed to allocate memory for label %s\n", buf); + fclose(file); + free_detect_labels(ctx); + return AVERROR(ENOMEM); + } + + if (av_dynarray_add_nofree(&ctx->labels, &ctx->label_count, label) < 0) { + av_log(context, AV_LOG_ERROR, "failed to do av_dynarray_add\n"); + fclose(file); + free_detect_labels(ctx); + return AVERROR(ENOMEM); + } + } + + fclose(file); + return 0; +} + +static av_cold int dnn_detect_init(AVFilterContext *context) +{ + DnnDetectContext *ctx = context->priv; + int ret = ff_dnn_init(&ctx->dnnctx, DFT_ANALYTICS_DETECT, context); + if (ret < 0) + return ret; + ff_dnn_set_proc(&ctx->dnnctx, NULL, dnn_detect_post_proc); + + if (ctx->labels_filename) { + return read_detect_label_file(context); + } + return 0; +} + +static int dnn_detect_query_formats(AVFilterContext *context) +{ + static const enum AVPixelFormat pix_fmts[] = { + AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24, + AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAYF32, + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P, + AV_PIX_FMT_NV12, + AV_PIX_FMT_NONE + }; + AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); + return ff_set_common_formats(context, fmts_list); +} + +static int dnn_detect_filter_frame(AVFilterLink *inlink, AVFrame *in) +{ + AVFilterContext *context = inlink->dst; + AVFilterLink *outlink = context->outputs[0]; + DnnDetectContext *ctx = context->priv; + DNNReturnType dnn_result; + + dnn_result = ff_dnn_execute_model(&ctx->dnnctx, in, NULL); + if (dnn_result != DNN_SUCCESS){ + av_log(ctx, AV_LOG_ERROR, "failed to execute model\n"); + av_frame_free(&in); + return AVERROR(EIO); + } + + return ff_filter_frame(outlink, in); +} + +static int dnn_detect_activate_sync(AVFilterContext *filter_ctx) +{ + AVFilterLink *inlink = filter_ctx->inputs[0]; + AVFilterLink *outlink = filter_ctx->outputs[0]; + AVFrame *in = NULL; + int64_t pts; + int ret, status; + int got_frame = 0; + + FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink); + + do { + // drain all input frames + ret = ff_inlink_consume_frame(inlink, &in); + if (ret < 0) + return ret; + if (ret > 0) { + ret = dnn_detect_filter_frame(inlink, in); + if (ret < 0) + return ret; + got_frame = 1; + } + } while (ret > 0); + + // if frame got, schedule to next filter + if (got_frame) + return 0; + + if (ff_inlink_acknowledge_status(inlink, &status, &pts)) { + if (status == AVERROR_EOF) { + ff_outlink_set_status(outlink, status, pts); + return ret; + } + } + + FF_FILTER_FORWARD_WANTED(outlink, inlink); + + return FFERROR_NOT_READY; +} + +static int dnn_detect_flush_frame(AVFilterLink *outlink, int64_t pts, int64_t *out_pts) +{ + DnnDetectContext *ctx = outlink->src->priv; + int ret; + DNNAsyncStatusType async_state; + + ret = ff_dnn_flush(&ctx->dnnctx); + if (ret != DNN_SUCCESS) { + return -1; + } + + do { + AVFrame *in_frame = NULL; + AVFrame *out_frame = NULL; + async_state = ff_dnn_get_async_result(&ctx->dnnctx, &in_frame, &out_frame); + if (out_frame) { + av_assert0(in_frame == out_frame); + ret = ff_filter_frame(outlink, out_frame); + if (ret < 0) + return ret; + if (out_pts) + *out_pts = out_frame->pts + pts; + } + av_usleep(5000); + } while (async_state >= DAST_NOT_READY); + + return 0; +} + +static int dnn_detect_activate_async(AVFilterContext *filter_ctx) +{ + AVFilterLink *inlink = filter_ctx->inputs[0]; + AVFilterLink *outlink = filter_ctx->outputs[0]; + DnnDetectContext *ctx = filter_ctx->priv; + AVFrame *in = NULL; + int64_t pts; + int ret, status; + int got_frame = 0; + int async_state; + + FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink); + + do { + // drain all input frames + ret = ff_inlink_consume_frame(inlink, &in); + if (ret < 0) + return ret; + if (ret > 0) { + if (ff_dnn_execute_model_async(&ctx->dnnctx, in, NULL) != DNN_SUCCESS) { + return AVERROR(EIO); + } + } + } while (ret > 0); + + // drain all processed frames + do { + AVFrame *in_frame = NULL; + AVFrame *out_frame = NULL; + async_state = ff_dnn_get_async_result(&ctx->dnnctx, &in_frame, &out_frame); + if (out_frame) { + av_assert0(in_frame == out_frame); + ret = ff_filter_frame(outlink, out_frame); + if (ret < 0) + return ret; + got_frame = 1; + } + } while (async_state == DAST_SUCCESS); + + // if frame got, schedule to next filter + if (got_frame) + return 0; + + if (ff_inlink_acknowledge_status(inlink, &status, &pts)) { + if (status == AVERROR_EOF) { + int64_t out_pts = pts; + ret = dnn_detect_flush_frame(outlink, pts, &out_pts); + ff_outlink_set_status(outlink, status, out_pts); + return ret; + } + } + + FF_FILTER_FORWARD_WANTED(outlink, inlink); + + return 0; +} + +static int dnn_detect_activate(AVFilterContext *filter_ctx) +{ + DnnDetectContext *ctx = filter_ctx->priv; + + if (ctx->dnnctx.async) + return dnn_detect_activate_async(filter_ctx); + else + return dnn_detect_activate_sync(filter_ctx); +} + +static av_cold void dnn_detect_uninit(AVFilterContext *context) +{ + DnnDetectContext *ctx = context->priv; + ff_dnn_uninit(&ctx->dnnctx); + free_detect_labels(ctx); +} + +static const AVFilterPad dnn_detect_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + }, + { NULL } +}; + +static const AVFilterPad dnn_detect_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + }, + { NULL } +}; + +AVFilter ff_vf_dnn_detect = { + .name = "dnn_detect", + .description = NULL_IF_CONFIG_SMALL("Apply DNN detect filter to the input."), + .priv_size = sizeof(DnnDetectContext), + .init = dnn_detect_init, + .uninit = dnn_detect_uninit, + .query_formats = dnn_detect_query_formats, + .inputs = dnn_detect_inputs, + .outputs = dnn_detect_outputs, + .priv_class = &dnn_detect_class, + .activate = dnn_detect_activate, +};