dnn/vf_dnn_detect.c: add tensorflow output parse support

Testing model is tensorflow offical model in github repo, please refer
https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md
to download the detect model as you need.
For example, local testing was carried on with 'ssd_mobilenet_v2_coco_2018_03_29.tar.gz', and
used one image of dog in
https://github.com/tensorflow/models/blob/master/research/object_detection/test_images/image1.jpg

Testing command is:
./ffmpeg -i image1.jpg -vf dnn_detect=dnn_backend=tensorflow:input=image_tensor:output=\
"num_detections&detection_scores&detection_classes&detection_boxes":model=ssd_mobilenet_v2_coco.pb,\
showinfo -f null -

We will see the result similar as below:
[Parsed_showinfo_1 @ 0x33e65f0]   side data - detection bounding boxes:
[Parsed_showinfo_1 @ 0x33e65f0] source: ssd_mobilenet_v2_coco.pb
[Parsed_showinfo_1 @ 0x33e65f0] index: 0,       region: (382, 60) -> (1005, 593), label: 18, confidence: 9834/10000.
[Parsed_showinfo_1 @ 0x33e65f0] index: 1,       region: (12, 8) -> (328, 549), label: 18, confidence: 8555/10000.
[Parsed_showinfo_1 @ 0x33e65f0] index: 2,       region: (293, 7) -> (682, 458), label: 1, confidence: 8033/10000.
[Parsed_showinfo_1 @ 0x33e65f0] index: 3,       region: (342, 0) -> (690, 325), label: 1, confidence: 5878/10000.

There are two boxes of dog with cores 94.05% & 93.45% and two boxes of person with scores 80.33% & 58.78%.

Signed-off-by: Ting Fu <ting.fu@intel.com>
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>

1 files changed, 94 insertions, 1 deletions

diff --git a/libavfilter/vf_dnn_detect.c b/libavfilter/vf_dnn_detect.c
index 7d39acb653..d23e30aedd 100644
--- a/libavfilter/vf_dnn_detect.c
+++ b/libavfilter/vf_dnn_detect.c
@@ -48,6 +48,9 @@ typedef struct DnnDetectContext {
 #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_LIBTENSORFLOW == 1)
+    { "tensorflow",  "tensorflow backend flag",    0,                        AV_OPT_TYPE_CONST,     { .i64 = 1 },    0, 0, FLAGS, "backend" },
+#endif
 #if (CONFIG_LIBOPENVINO == 1)
     { "openvino",    "openvino backend flag",      0,                        AV_OPT_TYPE_CONST,     { .i64 = 2 },    0, 0, FLAGS, "backend" },
 #endif
@@ -59,7 +62,7 @@ static const AVOption dnn_detect_options[] = {
 
 AVFILTER_DEFINE_CLASS(dnn_detect);
 
-static int dnn_detect_post_proc(AVFrame *frame, DNNData *output, uint32_t nb, AVFilterContext *filter_ctx)
+static int dnn_detect_post_proc_ov(AVFrame *frame, DNNData *output, AVFilterContext *filter_ctx)
 {
     DnnDetectContext *ctx = filter_ctx->priv;
     float conf_threshold = ctx->confidence;
@@ -136,6 +139,96 @@ static int dnn_detect_post_proc(AVFrame *frame, DNNData *output, uint32_t nb, AV
     return 0;
 }
 
+static int dnn_detect_post_proc_tf(AVFrame *frame, DNNData *output, AVFilterContext *filter_ctx)
+{
+    DnnDetectContext *ctx = filter_ctx->priv;
+    int proposal_count;
+    float conf_threshold = ctx->confidence;
+    float *conf, *position, *label_id, x0, y0, x1, y1;
+    int nb_bboxes = 0;
+    AVFrameSideData *sd;
+    AVDetectionBBox *bbox;
+    AVDetectionBBoxHeader *header;
+
+    proposal_count = *(float *)(output[0].data);
+    conf           = output[1].data;
+    position       = output[3].data;
+    label_id       = output[2].data;
+
+    sd = av_frame_get_side_data(frame, AV_FRAME_DATA_DETECTION_BBOXES);
+    if (sd) {
+        av_log(filter_ctx, AV_LOG_ERROR, "already have dnn bounding boxes in side data.\n");
+        return -1;
+    }
+
+    for (int i = 0; i < proposal_count; ++i) {
+        if (conf[i] < conf_threshold)
+            continue;
+        nb_bboxes++;
+    }
+
+    if (nb_bboxes == 0) {
+        av_log(filter_ctx, AV_LOG_VERBOSE, "nothing detected in this frame.\n");
+        return 0;
+    }
+
+    header = av_detection_bbox_create_side_data(frame, nb_bboxes);
+    if (!header) {
+        av_log(filter_ctx, AV_LOG_ERROR, "failed to create side data with %d bounding boxes\n", nb_bboxes);
+        return -1;
+    }
+
+    av_strlcpy(header->source, ctx->dnnctx.model_filename, sizeof(header->source));
+
+    for (int i = 0; i < proposal_count; ++i) {
+        y0 = position[i * 4];
+        x0 = position[i * 4 + 1];
+        y1 = position[i * 4 + 2];
+        x1 = position[i * 4 + 3];
+
+        bbox = av_get_detection_bbox(header, i);
+
+        if (conf[i] < conf_threshold) {
+            continue;
+        }
+
+        bbox->x = (int)(x0 * frame->width);
+        bbox->w = (int)(x1 * frame->width) - bbox->x;
+        bbox->y = (int)(y0 * frame->height);
+        bbox->h = (int)(y1 * frame->height) - bbox->y;
+
+        bbox->detect_confidence = av_make_q((int)(conf[i] * 10000), 10000);
+        bbox->classify_count = 0;
+
+        if (ctx->labels && label_id[i] < ctx->label_count) {
+            av_strlcpy(bbox->detect_label, ctx->labels[(int)label_id[i]], sizeof(bbox->detect_label));
+        } else {
+            snprintf(bbox->detect_label, sizeof(bbox->detect_label), "%d", (int)label_id[i]);
+        }
+
+        nb_bboxes--;
+        if (nb_bboxes == 0) {
+            break;
+        }
+    }
+    return 0;
+}
+
+static int dnn_detect_post_proc(AVFrame *frame, DNNData *output, uint32_t nb, AVFilterContext *filter_ctx)
+{
+    DnnDetectContext *ctx = filter_ctx->priv;
+    DnnContext *dnn_ctx = &ctx->dnnctx;
+    switch (dnn_ctx->backend_type) {
+    case DNN_OV:
+        return dnn_detect_post_proc_ov(frame, output, filter_ctx);
+    case DNN_TF:
+        return dnn_detect_post_proc_tf(frame, output, filter_ctx);
+    default:
+        avpriv_report_missing_feature(filter_ctx, "Current dnn backend does not support detect filter\n");
+        return AVERROR(EINVAL);
+    }
+}
+
 static void free_detect_labels(DnnDetectContext *ctx)
 {
     for (int i = 0; i < ctx->label_count; i++) {