[vc_row row_type=”3″ blox_image=”21762″ align_center=”aligncenter” page_title=”page-title-x” blox_dark=”true” parallax_speed=”8″][vc_column]

---

---

[vc_column_text]

Scene based adaptive encoding

[/vc_column_text]

---

---

[/vc_column][/vc_row][vc_row][vc_column width=”1/4″]

---

---

[/vc_column][vc_column width=”1/2″]

---

[vc_column_text]Startup in the video streaming industry needed scene based adaptive encoding.[/vc_column_text]

---

[/vc_column][vc_column width=”1/4″][/vc_column][/vc_row][vc_row][vc_column width=”1/3″][vc_single_image image=”21763″ img_size=”550×350″][/vc_column][vc_column width=”1/3″][vc_single_image image=”21764″ img_size=”550×350″][/vc_column][vc_column width=”1/3″][vc_single_image image=”21765″ img_size=”550×350″][/vc_column][/vc_row][vc_row][vc_column]

---

[vc_column_text]

About this project

[/vc_column_text][vc_row_inner][vc_column_inner width=”1/2″][vc_column_text]

Project description:

The company needed near real-time automatic CRF / Bitrate prediction per scene with limited (CPU only) resources.

Solution:

Frame and optical flow based motion complexity clusterization per scene with deep CNN model trained on the customized complexity estimators generated from the different compression parameters of raw images in the pipeline with CRF classification and Bitrate regression models.

[/vc_column_text][/vc_column_inner][vc_column_inner width=”1/2″][vc_column_text]

Technologies:

• PyTorch for deep CNN frame and motion complexity clusterization and CRF classification models and optical flow generation.
• Different OpenCV and FFmpeg based scene detection and Optical Flow generation techniques.
• XGBoost for Bitrate regression.

[/vc_column_text][/vc_column_inner][/vc_row_inner]

---

[/vc_column][/vc_row]