diff mbox series

[FFmpeg-devel,V3,3/3] dnn/openvino: support model input resize

Message ID 20210118034214.13116-3-ting.fu@intel.com
State Accepted
Headers show
Series [FFmpeg-devel,V3,1/3] dnn/openvino: remove unnecessary code
Related show

Checks

Context Check Description
andriy/x86_make success Make finished
andriy/x86_make_fate success Make fate finished
andriy/PPC64_make success Make finished
andriy/PPC64_make_fate success Make fate finished

Commit Message

Fu, Ting Jan. 18, 2021, 3:42 a.m. UTC
OpenVINO APIs require specify input size to run the model, while some
OpenVINO model does accept different input size. To enable this feature
adding input_resizable option here for easier use.
Setting bool variable input_resizable to specify if the input can be resizable or not.
input_resizable = 1 means support input resize, aka accept different input size.
input_resizable = 0 (default) means do not support input resize.
Please make sure the inference model does accept different input size
before use this option, otherwise the inference engine may report error(s).
eg: ./ffmpeg -i video_name.mp4 -vf dnn_processing=dnn_backend=openvino:\
      model=model_name.xml:input=input_name:output=output_name:\
      options=device=CPU\&input_resizable=1 -y output_video_name.mp4

Signed-off-by: Ting Fu <ting.fu@intel.com>
---
V3:
    rebase to latest code and add missing code

 libavfilter/dnn/dnn_backend_openvino.c | 21 +++++++++++++++++++--
 1 file changed, 19 insertions(+), 2 deletions(-)

Comments

Guo, Yejun Jan. 18, 2021, 5:30 a.m. UTC | #1
> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-bounces@ffmpeg.org> On Behalf Of Ting Fu
> Sent: 2021年1月18日 11:42
> To: ffmpeg-devel@ffmpeg.org
> Subject: [FFmpeg-devel] [PATCH V3 3/3] dnn/openvino: support model input
> resize
> 
> OpenVINO APIs require specify input size to run the model, while some
> OpenVINO model does accept different input size. To enable this feature
> adding input_resizable option here for easier use.
> Setting bool variable input_resizable to specify if the input can be resizable or
> not.
> input_resizable = 1 means support input resize, aka accept different input size.
> input_resizable = 0 (default) means do not support input resize.
> Please make sure the inference model does accept different input size before
> use this option, otherwise the inference engine may report error(s).
> eg: ./ffmpeg -i video_name.mp4 -vf dnn_processing=dnn_backend=openvino:\
>       model=model_name.xml:input=input_name:output=output_name:\
>       options=device=CPU\&input_resizable=1 -y output_video_name.mp4
> 
> Signed-off-by: Ting Fu <ting.fu@intel.com>
> ---
> V3:
>     rebase to latest code and add missing code
> 
>  libavfilter/dnn/dnn_backend_openvino.c | 21 +++++++++++++++++++--
>  1 file changed, 19 insertions(+), 2 deletions(-)
> 
this patch set looks good to me, will push soon, thanks.
diff mbox series

Patch

diff --git a/libavfilter/dnn/dnn_backend_openvino.c b/libavfilter/dnn/dnn_backend_openvino.c
index ecfd2b3f36..8a7abb33f0 100644
--- a/libavfilter/dnn/dnn_backend_openvino.c
+++ b/libavfilter/dnn/dnn_backend_openvino.c
@@ -38,6 +38,7 @@  typedef struct OVOptions{
     char *device_type;
     int nireq;
     int batch_size;
+    int input_resizable;
 } OVOptions;
 
 typedef struct OVContext {
@@ -86,6 +87,7 @@  static const AVOption dnn_openvino_options[] = {
     { "device", "device to run model", OFFSET(options.device_type), AV_OPT_TYPE_STRING, { .str = "CPU" }, 0, 0, FLAGS },
     { "nireq",  "number of request",   OFFSET(options.nireq),       AV_OPT_TYPE_INT,    { .i64 = 0 },     0, INT_MAX, FLAGS },
     { "batch_size",  "batch size per request", OFFSET(options.batch_size),  AV_OPT_TYPE_INT,    { .i64 = 1 },     1, 1000, FLAGS},
+    { "input_resizable", "can input be resizable or not", OFFSET(options.input_resizable), AV_OPT_TYPE_BOOL,   { .i64 = 0 },     0, 1, FLAGS },
     { NULL }
 };
 
@@ -400,6 +402,7 @@  static DNNReturnType get_input_ov(void *model, DNNData *input, const char *input
     size_t model_input_count = 0;
     dimensions_t dims;
     precision_e precision;
+    int input_resizable = ctx->options.input_resizable;
 
     status = ie_network_get_inputs_number(ov_model->network, &model_input_count);
     if (status != OK) {
@@ -423,8 +426,8 @@  static DNNReturnType get_input_ov(void *model, DNNData *input, const char *input
             }
 
             input->channels = dims.dims[1];
-            input->height   = dims.dims[2];
-            input->width    = dims.dims[3];
+            input->height   = input_resizable ? -1 : dims.dims[2];
+            input->width    = input_resizable ? -1 : dims.dims[3];
             input->dt       = precision_to_datatype(precision);
             return DNN_SUCCESS;
         } else {
@@ -450,6 +453,8 @@  static DNNReturnType get_output_ov(void *model, const char *input_name, int inpu
     AVFrame *in_frame = av_frame_alloc();
     AVFrame *out_frame = NULL;
     TaskItem *ptask = &task;
+    IEStatusCode status;
+    input_shapes_t input_shapes;
 
     if (!in_frame) {
         av_log(ctx, AV_LOG_ERROR, "Failed to allocate memory for input frame\n");
@@ -464,6 +469,18 @@  static DNNReturnType get_output_ov(void *model, const char *input_name, int inpu
     in_frame->width = input_width;
     in_frame->height = input_height;
 
+    if (ctx->options.input_resizable) {
+        status = ie_network_get_input_shapes(ov_model->network, &input_shapes);
+        input_shapes.shapes->shape.dims[2] = input_height;
+        input_shapes.shapes->shape.dims[3] = input_width;
+        status |= ie_network_reshape(ov_model->network, input_shapes);
+        ie_network_input_shapes_free(&input_shapes);
+        if (status != OK) {
+            av_log(ctx, AV_LOG_ERROR, "Failed to reshape input size for %s\n", input_name);
+            return DNN_ERROR;
+        }
+    }
+
     if (!ov_model->exe_network) {
         if (init_model_ov(ov_model) != DNN_SUCCESS) {
             av_log(ctx, AV_LOG_ERROR, "Failed init OpenVINO exectuable network or inference request\n");