User Interface#
To open the application the GUI, go to http://localhost:8080 in your browser.
You can change the address with make up UI_HOST_PORT=9090 and print the LAN URL with
make up and make run to be able to open the app from a different computer in the same
network.
All on-screen data is driven by a single Server-Sent Events stream at /api/events
(~1 Hz snapshot). There is no client-side state polling. The rendered video appears in a
native pop-up sink launched by the pipeline container when display mode is enabled.
Here is a detailed description of the layout:
Left column
Block |
Source |
Notes |
|---|---|---|
Source section |
local form state → |
Select |
Device section |
local form state → |
Select runtime target ( |
Session controls |
|
Start/Stop/Reset from the accordion instead of toolbar/modal flow. |
Right column — Pipeline Performance accordion
Column |
Source |
Meaning |
|---|---|---|
Workload |
static |
|
Model |
static |
|
Device |
|
Colored pill: |
FPS |
|
Rolling mean over the last ~5 s |
Mean |
|
Rolling mean pipeline latency from GST tracer samples |
P50 |
|
Median pipeline latency |
P90 |
|
90th percentile pipeline latency |
P95 |
|
95th percentile pipeline latency |
P99 |
|
99th percentile pipeline latency |
Status |
lifecycle FSM |
|
Below the table:
End-to-end summary bar — pipeline FPS · sample count · uptime · source kind.
Model & Input block — model name, precision (
FP16 OpenVINO IR), task/dataset (Polyp DetectiononColonDB), video source resolution (e.g.1080p H.264 (looped)), model input tensor size (640x640), and the runtime device.
Right column — Platform accordion
Live CPU / GPU / NPU utilization from intel-npu-info and nvidia-smi-style samplers,
refreshed on every SSE snapshot.