Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

libavfilter/dnn_interface: use dims to represent shapes

Browse Source 
For detect and classify output, width and height make no sence, so
change width, height to dims to represent the shape of tensor. Use
layout and dims to get width, height and channel.

Signed-off-by: Wenbin Chen <wenbin.chen@intel.com>
Reviewed-by: Guo Yejun <yejun.guo@intel.com>
release/7.0
Wenbin Chen 1 year ago committed by Guo Yejun
parent c695de56b5
commit 3de38b9da5
7 changed files with 146 additions and 90 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 80
      libavfilter/dnn/dnn_backend_openvino.c
  2. 32
      libavfilter/dnn/dnn_backend_tf.c
  3. 30
      libavfilter/dnn/dnn_io_proc.c
  4. 17
      libavfilter/dnn_interface.h
  5. 6
      libavfilter/vf_dnn_classify.c
  6. 50
      libavfilter/vf_dnn_detect.c
  7. 21
      libavfilter/vf_dnn_processing.c

80
libavfilter/dnn/dnn_backend_openvino.c
Unescape View File

@ -253,9 +253,9 @@ static int fill_model_input_ov(OVModel *ov_model, OVRequestItem *request)
ov_shape_free(&input_shape);
return ov2_map_error(status, NULL);
}
input.height = dims[1];
input.width = dims[2];
input.channels = dims[3];
for (int i = 0; i < input_shape.rank; i++)
input.dims[i] = dims[i];
input.layout = DL_NHWC;
input.dt = precision_to_datatype(precision);
#else
status = ie_infer_request_get_blob(request->infer_request, task->input_name, &input_blob);
@ -278,9 +278,9 @@ static int fill_model_input_ov(OVModel *ov_model, OVRequestItem *request)
av_log(ctx, AV_LOG_ERROR, "Failed to get input blob buffer\n");
return DNN_GENERIC_ERROR;
}
input.height = dims.dims[2];
input.width = dims.dims[3];
input.channels = dims.dims[1];
for (int i = 0; i < input_shape.rank; i++)
input.dims[i] = dims[i];
input.layout = DL_NCHW;
input.data = blob_buffer.buffer;
input.dt = precision_to_datatype(precision);
#endif
@ -339,8 +339,8 @@ static int fill_model_input_ov(OVModel *ov_model, OVRequestItem *request)
av_assert0(!"should not reach here");
break;
}
input.data = (uint8_t *)input.data
+ input.width * input.height * input.channels * get_datatype_size(input.dt);
input.data = (uint8_t *)input.data +
input.dims[1] * input.dims[2] * input.dims[3] * get_datatype_size(input.dt);
}
#if HAVE_OPENVINO2
ov_tensor_free(tensor);
@ -403,10 +403,11 @@ static void infer_completion_callback(void *args)
goto end;
}
outputs[i].dt = precision_to_datatype(precision);
outputs[i].channels = output_shape.rank > 2 ? dims[output_shape.rank - 3] : 1;
outputs[i].height = output_shape.rank > 1 ? dims[output_shape.rank - 2] : 1;
outputs[i].width = output_shape.rank > 0 ? dims[output_shape.rank - 1] : 1;
outputs[i].layout = DL_NCHW;
outputs[i].dims[0] = 1;
outputs[i].dims[1] = output_shape.rank > 2 ? dims[output_shape.rank - 3] : 1;
outputs[i].dims[2] = output_shape.rank > 1 ? dims[output_shape.rank - 2] : 1;
outputs[i].dims[3] = output_shape.rank > 0 ? dims[output_shape.rank - 1] : 1;
av_assert0(request->lltask_count <= dims[0]);
outputs[i].layout = ctx->options.layout;
outputs[i].scale = ctx->options.scale;
@ -445,9 +446,9 @@ static void infer_completion_callback(void *args)
return;
}
output.data = blob_buffer.buffer;
output.channels = dims.dims[1];
output.height = dims.dims[2];
output.width = dims.dims[3];
output.layout = DL_NCHW;
for (int i = 0; i < 4; i++)
output.dims[i] = dims.dims[i];
av_assert0(request->lltask_count <= dims.dims[0]);
output.dt = precision_to_datatype(precision);
output.layout = ctx->options.layout;
@ -469,8 +470,10 @@ static void infer_completion_callback(void *args)
ff_proc_from_dnn_to_frame(task->out_frame, outputs, ctx);
}
} else {
task->out_frame->width = outputs[0].width;
task->out_frame->height = outputs[0].height;
task->out_frame->width =
outputs[0].dims[dnn_get_width_idx_by_layout(outputs[0].layout)];
task->out_frame->height =
outputs[0].dims[dnn_get_height_idx_by_layout(outputs[0].layout)];
}
break;
case DFT_ANALYTICS_DETECT:
@ -501,7 +504,8 @@ static void infer_completion_callback(void *args)
av_freep(&request->lltasks[i]);
for (int i = 0; i < ov_model->nb_outputs; i++)
outputs[i].data = (uint8_t *)outputs[i].data +
outputs[i].width * outputs[i].height * outputs[i].channels * get_datatype_size(outputs[i].dt);
outputs[i].dims[1] * outputs[i].dims[2] * outputs[i].dims[3] *
get_datatype_size(outputs[i].dt);
}
end:
#if HAVE_OPENVINO2
@ -1085,7 +1089,6 @@ static int get_input_ov(void *model, DNNData *input, const char *input_name)
#if HAVE_OPENVINO2
ov_shape_t input_shape = {0};
ov_element_type_e precision;
int64_t* dims;
ov_status_e status;
if (input_name)
status = ov_model_const_input_by_name(ov_model->ov_model, input_name, &ov_model->input_port);
@ -1105,16 +1108,18 @@ static int get_input_ov(void *model, DNNData *input, const char *input_name)
av_log(ctx, AV_LOG_ERROR, "Failed to get input port shape.\n");
return ov2_map_error(status, NULL);
}
dims = input_shape.dims;
if (dims[1] <= 3) { // NCHW
input->channels = dims[1];
input->height = input_resizable ? -1 : dims[2];
input->width = input_resizable ? -1 : dims[3];
} else { // NHWC
input->height = input_resizable ? -1 : dims[1];
input->width = input_resizable ? -1 : dims[2];
input->channels = dims[3];
for (int i = 0; i < 4; i++)
input->dims[i] = input_shape.dims[i];
if (input_resizable) {
input->dims[dnn_get_width_idx_by_layout(input->layout)] = -1;
input->dims[dnn_get_height_idx_by_layout(input->layout)] = -1;
}
if (input_shape.dims[1] <= 3) // NCHW
input->layout = DL_NCHW;
else // NHWC
input->layout = DL_NHWC;
input->dt = precision_to_datatype(precision);
ov_shape_free(&input_shape);
return 0;
@ -1144,15 +1149,18 @@ static int get_input_ov(void *model, DNNData *input, const char *input_name)
return DNN_GENERIC_ERROR;
}
if (dims[1] <= 3) { // NCHW
input->channels = dims[1];
input->height = input_resizable ? -1 : dims[2];
input->width = input_resizable ? -1 : dims[3];
} else { // NHWC
input->height = input_resizable ? -1 : dims[1];
input->width = input_resizable ? -1 : dims[2];
input->channels = dims[3];
for (int i = 0; i < 4; i++)
input->dims[i] = input_shape.dims[i];
if (input_resizable) {
input->dims[dnn_get_width_idx_by_layout(input->layout)] = -1;
input->dims[dnn_get_height_idx_by_layout(input->layout)] = -1;
}
if (input_shape.dims[1] <= 3) // NCHW
input->layout = DL_NCHW;
else // NHWC
input->layout = DL_NHWC;
input->dt = precision_to_datatype(precision);
return 0;
}

32
libavfilter/dnn/dnn_backend_tf.c
Unescape View File

@ -251,7 +251,12 @@ static TF_Tensor *allocate_input_tensor(const DNNData *input)
{
TF_DataType dt;
size_t size;
int64_t input_dims[] = {1, input->height, input->width, input->channels};
int64_t input_dims[4] = { 0 };
input_dims[0] = 1;
input_dims[1] = input->dims[dnn_get_height_idx_by_layout(input->layout)];
input_dims[2] = input->dims[dnn_get_width_idx_by_layout(input->layout)];
input_dims[3] = input->dims[dnn_get_channel_idx_by_layout(input->layout)];
switch (input->dt) {
case DNN_FLOAT:
dt = TF_FLOAT;
@ -310,9 +315,9 @@ static int get_input_tf(void *model, DNNData *input, const char *input_name)
// currently only NHWC is supported
av_assert0(dims[0] == 1 || dims[0] == -1);
input->height = dims[1];
input->width = dims[2];
input->channels = dims[3];
for (int i = 0; i < 4; i++)
input->dims[i] = dims[i];
input->layout = DL_NHWC;
return 0;
}
@ -640,8 +645,8 @@ static int fill_model_input_tf(TFModel *tf_model, TFRequestItem *request) {
}
infer_request = request->infer_request;
input.height = task->in_frame->height;
input.width = task->in_frame->width;
input.dims[1] = task->in_frame->height;
input.dims[2] = task->in_frame->width;
infer_request->tf_input = av_malloc(sizeof(TF_Output));
if (!infer_request->tf_input) {
@ -731,9 +736,12 @@ static void infer_completion_callback(void *args) {
}
for (uint32_t i = 0; i < task->nb_output; ++i) {
outputs[i].height = TF_Dim(infer_request->output_tensors[i], 1);
outputs[i].width = TF_Dim(infer_request->output_tensors[i], 2);
outputs[i].channels = TF_Dim(infer_request->output_tensors[i], 3);
outputs[i].dims[dnn_get_height_idx_by_layout(outputs[i].layout)] =
TF_Dim(infer_request->output_tensors[i], 1);
outputs[i].dims[dnn_get_width_idx_by_layout(outputs[i].layout)] =
TF_Dim(infer_request->output_tensors[i], 2);
outputs[i].dims[dnn_get_channel_idx_by_layout(outputs[i].layout)] =
TF_Dim(infer_request->output_tensors[i], 3);
outputs[i].data = TF_TensorData(infer_request->output_tensors[i]);
outputs[i].dt = (DNNDataType)TF_TensorType(infer_request->output_tensors[i]);
}
@ -747,8 +755,10 @@ static void infer_completion_callback(void *args) {
ff_proc_from_dnn_to_frame(task->out_frame, outputs, ctx);
}
} else {
task->out_frame->width = outputs[0].width;
task->out_frame->height = outputs[0].height;
task->out_frame->width =
outputs[0].dims[dnn_get_width_idx_by_layout(outputs[0].layout)];
task->out_frame->height =
outputs[0].dims[dnn_get_height_idx_by_layout(outputs[0].layout)];
}
break;
case DFT_ANALYTICS_DETECT:

30
libavfilter/dnn/dnn_io_proc.c
Unescape View File

@ -70,7 +70,7 @@ int ff_proc_from_dnn_to_frame(AVFrame *frame, DNNData *output, void *log_ctx)
dst_data = (void **)frame->data;
linesize[0] = frame->linesize[0];
if (output->layout == DL_NCHW) {
middle_data = av_malloc(plane_size * output->channels);
middle_data = av_malloc(plane_size * output->dims[1]);
if (!middle_data) {
ret = AVERROR(ENOMEM);
goto err;
@ -209,7 +209,7 @@ int ff_proc_from_frame_to_dnn(AVFrame *frame, DNNData *input, void *log_ctx)
src_data = (void **)frame->data;
linesize[0] = frame->linesize[0];
if (input->layout == DL_NCHW) {
middle_data = av_malloc(plane_size * input->channels);
middle_data = av_malloc(plane_size * input->dims[1]);
if (!middle_data) {
ret = AVERROR(ENOMEM);
goto err;
@ -346,6 +346,7 @@ int ff_frame_to_dnn_classify(AVFrame *frame, DNNData *input, uint32_t bbox_index
int ret = 0;
enum AVPixelFormat fmt;
int left, top, width, height;
int width_idx, height_idx;
const AVDetectionBBoxHeader *header;
const AVDetectionBBox *bbox;
AVFrameSideData *sd = av_frame_get_side_data(frame, AV_FRAME_DATA_DETECTION_BBOXES);
@ -364,6 +365,9 @@ int ff_frame_to_dnn_classify(AVFrame *frame, DNNData *input, uint32_t bbox_index
return AVERROR(ENOSYS);
}
width_idx = dnn_get_width_idx_by_layout(input->layout);
height_idx = dnn_get_height_idx_by_layout(input->layout);
header = (const AVDetectionBBoxHeader *)sd->data;
bbox = av_get_detection_bbox(header, bbox_index);
@ -374,17 +378,20 @@ int ff_frame_to_dnn_classify(AVFrame *frame, DNNData *input, uint32_t bbox_index
fmt = get_pixel_format(input);
sws_ctx = sws_getContext(width, height, frame->format,
input->width, input->height, fmt,
input->dims[width_idx],
input->dims[height_idx], fmt,
SWS_FAST_BILINEAR, NULL, NULL, NULL);
if (!sws_ctx) {
av_log(log_ctx, AV_LOG_ERROR, "Failed to create scale context for the conversion "
"fmt:%s s:%dx%d -> fmt:%s s:%dx%d\n",
av_get_pix_fmt_name(frame->format), width, height,
av_get_pix_fmt_name(fmt), input->width, input->height);
av_get_pix_fmt_name(fmt),
input->dims[width_idx],
input->dims[height_idx]);
return AVERROR(EINVAL);
}
ret = av_image_fill_linesizes(linesizes, fmt, input->width);
ret = av_image_fill_linesizes(linesizes, fmt, input->dims[width_idx]);
if (ret < 0) {
av_log(log_ctx, AV_LOG_ERROR, "unable to get linesizes with av_image_fill_linesizes");
sws_freeContext(sws_ctx);
@ -414,7 +421,7 @@ int ff_frame_to_dnn_detect(AVFrame *frame, DNNData *input, void *log_ctx)
{
struct SwsContext *sws_ctx;
int linesizes[4];
int ret = 0;
int ret = 0, width_idx, height_idx;
enum AVPixelFormat fmt = get_pixel_format(input);
/* (scale != 1 and scale != 0) or mean != 0 */
@ -430,18 +437,23 @@ int ff_frame_to_dnn_detect(AVFrame *frame, DNNData *input, void *log_ctx)
return AVERROR(ENOSYS);
}
width_idx = dnn_get_width_idx_by_layout(input->layout);
height_idx = dnn_get_height_idx_by_layout(input->layout);
sws_ctx = sws_getContext(frame->width, frame->height, frame->format,
input->width, input->height, fmt,
input->dims[width_idx],
input->dims[height_idx], fmt,
SWS_FAST_BILINEAR, NULL, NULL, NULL);
if (!sws_ctx) {
av_log(log_ctx, AV_LOG_ERROR, "Impossible to create scale context for the conversion "
"fmt:%s s:%dx%d -> fmt:%s s:%dx%d\n",
av_get_pix_fmt_name(frame->format), frame->width, frame->height,
av_get_pix_fmt_name(fmt), input->width, input->height);
av_get_pix_fmt_name(fmt), input->dims[width_idx],
input->dims[height_idx]);
return AVERROR(EINVAL);
}
ret = av_image_fill_linesizes(linesizes, fmt, input->width);
ret = av_image_fill_linesizes(linesizes, fmt, input->dims[width_idx]);
if (ret < 0) {
av_log(log_ctx, AV_LOG_ERROR, "unable to get linesizes with av_image_fill_linesizes");
sws_freeContext(sws_ctx);

17
libavfilter/dnn_interface.h
Unescape View File

@ -64,7 +64,7 @@ typedef enum {
typedef struct DNNData{
void *data;
int width, height, channels;
int dims[4];
// dt and order together decide the color format
DNNDataType dt;
DNNColorOrder order;
@ -134,4 +134,19 @@ typedef struct DNNModule{
// Initializes DNNModule depending on chosen backend.
const DNNModule *ff_get_dnn_module(DNNBackendType backend_type, void *log_ctx);
static inline int dnn_get_width_idx_by_layout(DNNLayout layout)
{
return layout == DL_NHWC ? 2 : 3;
}
static inline int dnn_get_height_idx_by_layout(DNNLayout layout)
{
return layout == DL_NHWC ? 1 : 2;
}
static inline int dnn_get_channel_idx_by_layout(DNNLayout layout)
{
return layout == DL_NHWC ? 3 : 1;
}
#endif

6
libavfilter/vf_dnn_classify.c
Unescape View File

@ -68,8 +68,8 @@ static int dnn_classify_post_proc(AVFrame *frame, DNNData *output, uint32_t bbox
uint32_t label_id;
float confidence;
AVFrameSideData *sd;
if (output->channels <= 0) {
int output_size = output->dims[3] * output->dims[2] * output->dims[1];
if (output_size <= 0) {
return -1;
}
@ -88,7 +88,7 @@ static int dnn_classify_post_proc(AVFrame *frame, DNNData *output, uint32_t bbox
classifications = output->data;
label_id = 0;
confidence= classifications[0];
for (int i = 1; i < output->channels; i++) {
for (int i = 1; i < output_size; i++) {
if (classifications[i] > confidence) {
label_id = i;
confidence= classifications[i];

50
libavfilter/vf_dnn_detect.c
Unescape View File

@ -166,14 +166,14 @@ static int dnn_detect_parse_yolo_output(AVFrame *frame, DNNData *output, int out
scale_w = cell_w;
scale_h = cell_h;
} else {
if (output[output_index].height != output[output_index].width &&
output[output_index].height == output[output_index].channels) {
if (output[output_index].dims[2] != output[output_index].dims[3] &&
output[output_index].dims[2] == output[output_index].dims[1]) {
is_NHWC = 1;
cell_w = output[output_index].height;
cell_h = output[output_index].channels;
cell_w = output[output_index].dims[2];
cell_h = output[output_index].dims[1];
} else {
cell_w = output[output_index].width;
cell_h = output[output_index].height;
cell_w = output[output_index].dims[3];
cell_h = output[output_index].dims[2];
}
scale_w = ctx->scale_width;
scale_h = ctx->scale_height;
@ -205,14 +205,14 @@ static int dnn_detect_parse_yolo_output(AVFrame *frame, DNNData *output, int out
return AVERROR(EINVAL);
}
if (output[output_index].channels * output[output_index].width *
output[output_index].height % (box_size * cell_w * cell_h)) {
if (output[output_index].dims[1] * output[output_index].dims[2] *
output[output_index].dims[3] % (box_size * cell_w * cell_h)) {
av_log(filter_ctx, AV_LOG_ERROR, "wrong cell_w, cell_h or nb_classes\n");
return AVERROR(EINVAL);
}
detection_boxes = output[output_index].channels *
output[output_index].height *
output[output_index].width / box_size / cell_w / cell_h;
detection_boxes = output[output_index].dims[1] *
output[output_index].dims[2] *
output[output_index].dims[3] / box_size / cell_w / cell_h;
anchors = anchors + (detection_boxes * output_index * 2);
/**
@ -373,18 +373,18 @@ static int dnn_detect_post_proc_ssd(AVFrame *frame, DNNData *output, int nb_outp
int scale_w = ctx->scale_width;
int scale_h = ctx->scale_height;
if (nb_outputs == 1 && output->width == 7) {
proposal_count = output->height;
detect_size = output->width;
if (nb_outputs == 1 && output->dims[3] == 7) {
proposal_count = output->dims[2];
detect_size = output->dims[3];
detections = output->data;
} else if (nb_outputs == 2 && output[0].width == 5) {
proposal_count = output[0].height;
detect_size = output[0].width;
} else if (nb_outputs == 2 && output[0].dims[3] == 5) {
proposal_count = output[0].dims[2];
detect_size = output[0].dims[3];
detections = output[0].data;
labels = output[1].data;
} else if (nb_outputs == 2 && output[1].width == 5) {
proposal_count = output[1].height;
detect_size = output[1].width;
} else if (nb_outputs == 2 && output[1].dims[3] == 5) {
proposal_count = output[1].dims[2];
detect_size = output[1].dims[3];
detections = output[1].data;
labels = output[0].data;
} else {
@ -821,15 +821,19 @@ static int config_input(AVFilterLink *inlink)
AVFilterContext *context = inlink->dst;
DnnDetectContext *ctx = context->priv;
DNNData model_input;
int ret;
int ret, width_idx, height_idx;
ret = ff_dnn_get_input(&ctx->dnnctx, &model_input);
if (ret != 0) {
av_log(ctx, AV_LOG_ERROR, "could not get input from the model\n");
return ret;
}
ctx->scale_width = model_input.width == -1 ? inlink->w : model_input.width;
ctx->scale_height = model_input.height == -1 ? inlink->h : model_input.height;
width_idx = dnn_get_width_idx_by_layout(model_input.layout);
height_idx = dnn_get_height_idx_by_layout(model_input.layout);
ctx->scale_width = model_input.dims[width_idx] == -1 ? inlink->w :
model_input.dims[width_idx];
ctx->scale_height = model_input.dims[height_idx] == -1 ? inlink->h :
model_input.dims[height_idx];
return 0;
}

21
libavfilter/vf_dnn_processing.c
Unescape View File

@ -77,22 +77,29 @@ static const enum AVPixelFormat pix_fmts[] = {
"the frame's format %s does not match " \
"the model input channel %d\n", \
av_get_pix_fmt_name(fmt), \
model_input->channels);
model_input->dims[dnn_get_channel_idx_by_layout(model_input->layout)]);
static int check_modelinput_inlink(const DNNData *model_input, const AVFilterLink *inlink)
{
AVFilterContext *ctx = inlink->dst;
enum AVPixelFormat fmt = inlink->format;
int width_idx, height_idx;
width_idx = dnn_get_width_idx_by_layout(model_input->layout);
height_idx = dnn_get_height_idx_by_layout(model_input->layout);
// the design is to add explicit scale filter before this filter
if (model_input->height != -1 && model_input->height != inlink->h) {
if (model_input->dims[height_idx] != -1 &&
model_input->dims[height_idx] != inlink->h) {
av_log(ctx, AV_LOG_ERROR, "the model requires frame height %d but got %d\n",
model_input->height, inlink->h);
model_input->dims[height_idx],
inlink->h);
return AVERROR(EIO);
}
if (model_input->width != -1 && model_input->width != inlink->w) {
if (model_input->dims[width_idx] != -1 &&
model_input->dims[width_idx] != inlink->w) {
av_log(ctx, AV_LOG_ERROR, "the model requires frame width %d but got %d\n",
model_input->width, inlink->w);
model_input->dims[width_idx],
inlink->w);
return AVERROR(EIO);
}
if (model_input->dt != DNN_FLOAT) {
@ -103,7 +110,7 @@ static int check_modelinput_inlink(const DNNData *model_input, const AVFilterLin
switch (fmt) {
case AV_PIX_FMT_RGB24:
case AV_PIX_FMT_BGR24:
if (model_input->channels != 3) {
if (model_input->dims[dnn_get_channel_idx_by_layout(model_input->layout)] != 3) {
LOG_FORMAT_CHANNEL_MISMATCH();
return AVERROR(EIO);
}
@ -116,7 +123,7 @@ static int check_modelinput_inlink(const DNNData *model_input, const AVFilterLin
case AV_PIX_FMT_YUV410P:
case AV_PIX_FMT_YUV411P:
case AV_PIX_FMT_NV12:
if (model_input->channels != 1) {
if (model_input->dims[dnn_get_channel_idx_by_layout(model_input->layout)] != 1) {
LOG_FORMAT_CHANNEL_MISMATCH();
return AVERROR(EIO);
}

Write Preview
Loading…
Cancel
Save