Unified · Split
| @@ -136,40 +136,40 @@ static int config_input(AVFilterLink *inlink) · 135 unmodified lines | |||
| 136 | AVFilterContext *context = inlink->dst; | 136 | AVFilterContext *context = inlink->dst; |
| 137 | DnnProcessingContext *ctx = context->priv; | 137 | DnnProcessingContext *ctx = context->priv; |
| 138 | DNNReturnType result; | 138 | DNNReturnType result; |
| 139 | DNNData dnn_data; | 139 | DNNData model_input; |
| 140 | 140 | ||
| 141 | result = ctx->model->get_input(ctx->model->model, &dnn_data, ctx->model_inputname); | 141 | result = ctx->model->get_input(ctx->model->model, &model_input, ctx->model_inputname); |
| 142 | if (result != DNN_SUCCESS) { | 142 | if (result != DNN_SUCCESS) { |
| 143 | av_log(ctx, AV_LOG_ERROR, "could not get input from the model\n"); | 143 | av_log(ctx, AV_LOG_ERROR, "could not get input from the model\n"); |
| 144 | return AVERROR(EIO); | 144 | return AVERROR(EIO); |
| 145 | } | 145 | } |
| 146 | 146 | ||
| 147 | // the design is to add explicit scale filter before this filter | 147 | // the design is to add explicit scale filter before this filter |
| 148 | if (dnn_data.height != -1 && dnn_data.height != inlink->h) { | 148 | if (model_input.height != -1 && model_input.height != inlink->h) { |
| 149 | av_log(ctx, AV_LOG_ERROR, "the model requires frame height %d but got %d\n", | 149 | av_log(ctx, AV_LOG_ERROR, "the model requires frame height %d but got %d\n", |
| 150 | dnn_data.height, inlink->h); | 150 | model_input.height, inlink->h); |
| 151 | return AVERROR(EIO); | 151 | return AVERROR(EIO); |
| 152 | } | 152 | } |
| 153 | if (dnn_data.width != -1 && dnn_data.width != inlink->w) { | 153 | if (model_input.width != -1 && model_input.width != inlink->w) { |
| 154 | av_log(ctx, AV_LOG_ERROR, "the model requires frame width %d but got %d\n", | 154 | av_log(ctx, AV_LOG_ERROR, "the model requires frame width %d but got %d\n", |
| 155 | dnn_data.width, inlink->w); | 155 | model_input.width, inlink->w); |
| 156 | return AVERROR(EIO); | 156 | return AVERROR(EIO); |
| 157 | } | 157 | } |
| 158 | 158 | ||
| 159 | if (dnn_data.channels != 3) { | 159 | if (model_input.channels != 3) { |
| 160 | av_log(ctx, AV_LOG_ERROR, "the model requires input channels %d\n", | 160 | av_log(ctx, AV_LOG_ERROR, "the model requires input channels %d\n", |
| 161 | dnn_data.channels); | 161 | model_input.channels); |
| 162 | return AVERROR(EIO); | 162 | return AVERROR(EIO); |
| 163 | } | 163 | } |
| 164 | if (dnn_data.dt != DNN_FLOAT && dnn_data.dt != DNN_UINT8) { | 164 | if (model_input.dt != DNN_FLOAT && model_input.dt != DNN_UINT8) { |
| 165 | av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type as float32 and uint8.\n"); | 165 | av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type as float32 and uint8.\n"); |
| 166 | return AVERROR(EIO); | 166 | return AVERROR(EIO); |
| 167 | } | 167 | } |
| 168 | 168 | ||
| 169 | ctx->input.width = inlink->w; | 169 | ctx->input.width = inlink->w; |
| 170 | ctx->input.height = inlink->h; | 170 | ctx->input.height = inlink->h; |
| 171 | ctx->input.channels = dnn_data.channels; | 171 | ctx->input.channels = model_input.channels; |
| 172 | ctx->input.dt = dnn_data.dt; | 172 | ctx->input.dt = model_input.dt; |
| 173 | 173 | ||
| 174 | result = (ctx->model->set_input_output)(ctx->model->model, | 174 | result = (ctx->model->set_input_output)(ctx->model->model, |
| 175 | &ctx->input, ctx->model_inputname, | 175 | &ctx->input, ctx->model_inputname, |
| @@ -201,28 +201,28 @@ static int config_output(AVFilterLink *outlink) · 25 unmodified lines | |||
| 201 | return 0; | 201 | return 0; |
| 202 | } | 202 | } |
| 203 | 203 | ||
| 204 | static int copy_from_frame_to_dnn(DNNData *dnn_data, const AVFrame *in) | 204 | static int copy_from_frame_to_dnn(DNNData *dnn_input, const AVFrame *frame) |
| 205 | { | 205 | { |
| 206 | // extend this function to support more formats | 206 | // extend this function to support more formats |
| 207 | av_assert0(in->format == AV_PIX_FMT_RGB24 || in->format == AV_PIX_FMT_BGR24); | 207 | av_assert0(frame->format == AV_PIX_FMT_RGB24 || frame->format == AV_PIX_FMT_BGR24); |
| 208 | 208 | ||
| 209 | if (dnn_data->dt == DNN_FLOAT) { | 209 | if (dnn_input->dt == DNN_FLOAT) { |
| 210 | float *dnn_input = dnn_data->data; | 210 | float *dnn_input_data = dnn_input->data; |
| 211 | for (int i = 0; i < in->height; i++) { | 211 | for (int i = 0; i < frame->height; i++) { |
| 212 | for(int j = 0; j < in->width * 3; j++) { | 212 | for(int j = 0; j < frame->width * 3; j++) { |
| 213 | int k = i * in->linesize[0] + j; | 213 | int k = i * frame->linesize[0] + j; |
| 214 | int t = i * in->width * 3 + j; | 214 | int t = i * frame->width * 3 + j; |
| 215 | dnn_input[t] = in->data[0][k] / 255.0f; | 215 | dnn_input_data[t] = frame->data[0][k] / 255.0f; |
| 216 | } | 216 | } |
| 217 | } | 217 | } |
| 218 | } else { | 218 | } else { |
| 219 | uint8_t *dnn_input = dnn_data->data; | 219 | uint8_t *dnn_input_data = dnn_input->data; |
| 220 | av_assert0(dnn_data->dt == DNN_UINT8); | 220 | av_assert0(dnn_input->dt == DNN_UINT8); |
| 221 | for (int i = 0; i < in->height; i++) { | 221 | for (int i = 0; i < frame->height; i++) { |
| 222 | for(int j = 0; j < in->width * 3; j++) { | 222 | for(int j = 0; j < frame->width * 3; j++) { |
| 223 | int k = i * in->linesize[0] + j; | 223 | int k = i * frame->linesize[0] + j; |
| 224 | int t = i * in->width * 3 + j; | 224 | int t = i * frame->width * 3 + j; |
| 225 | dnn_input[t] = in->data[0][k]; | 225 | dnn_input_data[t] = frame->data[0][k]; |
| 226 | } | 226 | } |
| 227 | } | 227 | } |
| 228 | } | 228 | } |
| @@ -230,28 +230,28 @@ static int copy_from_frame_to_dnn(DNNData *dnn_data, const AVFrame *in) · 1 unmodified line | |||
| 230 | return 0; | 230 | return 0; |
| 231 | } | 231 | } |
| 232 | 232 | ||
| 233 | static int copy_from_dnn_to_frame(AVFrame *out, const DNNData *dnn_data) | 233 | static int copy_from_dnn_to_frame(AVFrame *frame, const DNNData *dnn_output) |
| 234 | { | 234 | { |
| 235 | // extend this function to support more formats | 235 | // extend this function to support more formats |
| 236 | av_assert0(out->format == AV_PIX_FMT_RGB24 || out->format == AV_PIX_FMT_BGR24); | 236 | av_assert0(frame->format == AV_PIX_FMT_RGB24 || frame->format == AV_PIX_FMT_BGR24); |
| 237 | 237 | ||
| 238 | if (dnn_data->dt == DNN_FLOAT) { | 238 | if (dnn_output->dt == DNN_FLOAT) { |
| 239 | float *dnn_output = dnn_data->data; | 239 | float *dnn_output_data = dnn_output->data; |
| 240 | for (int i = 0; i < out->height; i++) { | 240 | for (int i = 0; i < frame->height; i++) { |
| 241 | for(int j = 0; j < out->width * 3; j++) { | 241 | for(int j = 0; j < frame->width * 3; j++) { |
| 242 | int k = i * out->linesize[0] + j; | 242 | int k = i * frame->linesize[0] + j; |
| 243 | int t = i * out->width * 3 + j; | 243 | int t = i * frame->width * 3 + j; |
| 244 | out->data[0][k] = av_clip_uintp2((int)(dnn_output[t] * 255.0f), 8); | 244 | frame->data[0][k] = av_clip_uintp2((int)(dnn_output_data[t] * 255.0f), 8); |
| 245 | } | 245 | } |
| 246 | } | 246 | } |
| 247 | } else { | 247 | } else { |
| 248 | uint8_t *dnn_output = dnn_data->data; | 248 | uint8_t *dnn_output_data = dnn_output->data; |
| 249 | av_assert0(dnn_data->dt == DNN_UINT8); | 249 | av_assert0(dnn_output->dt == DNN_UINT8); |
| 250 | for (int i = 0; i < out->height; i++) { | 250 | for (int i = 0; i < frame->height; i++) { |
| 251 | for(int j = 0; j < out->width * 3; j++) { | 251 | for(int j = 0; j < frame->width * 3; j++) { |
| 252 | int k = i * out->linesize[0] + j; | 252 | int k = i * frame->linesize[0] + j; |
| 253 | int t = i * out->width * 3 + j; | 253 | int t = i * frame->width * 3 + j; |
| 254 | out->data[0][k] = dnn_output[t]; | 254 | frame->data[0][k] = dnn_output_data[t]; |
| 255 | } | 255 | } |
| 256 | } | 256 | } |
| 257 | } | 257 | } |