Getting slow FPS for Still images

AlanElkin · Jun 3, 2024

I've tried adding pipeline.setXLinkChunkSize(0) and results improved (thanks!), but FPS is still not what it should be: around 30 FPS. Anything else or any thought on this?

This is the new output:

Received 1 frames in 0:00:00.903939 --> Rate: 1.11 fps
Received 1 frames in 0:00:00.061003 --> Rate: 16.39 fps
Received 1 frames in 0:00:00.052928 --> Rate: 18.89 fps
Received 1 frames in 0:00:00.057465 --> Rate: 17.40 fps
Received 2 frames in 0:00:00.110694 --> Rate: 18.07 fps
Received 1 frames in 0:00:00.055616 --> Rate: 17.98 fps
Received 2 frames in 0:00:00.112372 --> Rate: 17.80 fps
Received 1 frames in 0:00:00.064235 --> Rate: 15.57 fps
Received 1 frames in 0:00:00.066208 --> Rate: 15.10 fps
Received 2 frames in 0:00:00.102976 --> Rate: 19.42 fps
Received 2 frames in 0:00:00.115193 --> Rate: 17.36 fps
Received 1 frames in 0:00:00.054441 --> Rate: 18.37 fps
Received 2 frames in 0:00:00.122025 --> Rate: 16.39 fps
Received 1 frames in 0:00:00.053625 --> Rate: 18.65 fps
...

jakaskerl · Jun 3, 2024

Hi @AlanElkin
Only thing I changed is the chunk size:

Received 1 frames in 0:00:00.033700 --> Rate: 29.67 fps
Received 1 frames in 0:00:00.034360 --> Rate: 29.10 fps
Received 1 frames in 0:00:00.033957 --> Rate: 29.45 fps
Received 1 frames in 0:00:00.031237 --> Rate: 32.01 fps
Received 1 frames in 0:00:00.034367 --> Rate: 29.10 fps
Received 1 frames in 0:00:00.033948 --> Rate: 29.46 fps
Received 1 frames in 0:00:00.032999 --> Rate: 30.30 fps
Received 1 frames in 0:00:00.034044 --> Rate: 29.37 fps
Received 1 frames in 0:00:00.032406 --> Rate: 30.86 fps
Received 1 frames in 0:00:00.034284 --> Rate: 29.17 fps
Received 1 frames in 0:00:00.032633 --> Rate: 30.64 fps
Received 1 frames in 0:00:00.033902 --> Rate: 29.50 fps
Received 1 frames in 0:00:00.034529 --> Rate: 28.96 fps
Received 1 frames in 0:00:00.032338 --> Rate: 30.92 fps
Received 1 frames in 0:00:00.033070 --> Rate: 30.24 fps
Received 1 frames in 0:00:00.033563 --> Rate: 29.79 fps

Are you using a POE device or a slow USB cable by any chance? luxonis/depthai-experimentsblob/master/random-scripts/oak_bandwidth_test.py

Thanks,
Jaka

AlanElkin · Jun 3, 2024

Hi jakaskerl and thanks again for your reply.

I'm using the provided USB cable from Luxonis and connecting my OAK1 directly to a USB 3 port in my laptop, with no PoE or anything else.

I've ran this test and got this:

Downlink 2134.4 mbps
Uplink 1995.6 mbps

I've also ran this test, which outputs:

UsbSpeed.SUPER
Latency: 38.18 ms, Average latency: 38.18 ms, Std: 0.00
Latency: 38.55 ms, Average latency: 38.37 ms, Std: 0.18
Latency: 36.84 ms, Average latency: 37.86 ms, Std: 0.73
Latency: 33.30 ms, Average latency: 36.72 ms, Std: 2.07
Latency: 33.05 ms, Average latency: 35.98 ms, Std: 2.36
Latency: 33.68 ms, Average latency: 35.60 ms, Std: 2.32
Latency: 33.78 ms, Average latency: 35.34 ms, Std: 2.24
Latency: 33.11 ms, Average latency: 35.06 ms, Std: 2.22
Latency: 33.29 ms, Average latency: 34.86 ms, Std: 2.17
Latency: 33.74 ms, Average latency: 34.75 ms, Std: 2.09
Latency: 33.35 ms, Average latency: 34.62 ms, Std: 2.03
Latency: 33.63 ms, Average latency: 34.54 ms, Std: 1.96
Latency: 33.40 ms, Average latency: 34.45 ms, Std: 1.91
Latency: 33.16 ms, Average latency: 34.36 ms, Std: 1.87
Latency: 33.56 ms, Average latency: 34.31 ms, Std: 1.82
Latency: 33.30 ms, Average latency: 34.24 ms, Std: 1.78
Latency: 33.14 ms, Average latency: 34.18 ms, Std: 1.74
Latency: 33.41 ms, Average latency: 34.14 ms, Std: 1.70
Latency: 33.25 ms, Average latency: 34.09 ms, Std: 1.67
Latency: 33.16 ms, Average latency: 34.04 ms, Std: 1.64
Latency: 33.72 ms, Average latency: 34.03 ms, Std: 1.60
Latency: 33.50 ms, Average latency: 34.00 ms, Std: 1.57
Latency: 34.27 ms, Average latency: 34.02 ms, Std: 1.54
Latency: 33.98 ms, Average latency: 34.01 ms, Std: 1.50
Latency: 33.65 ms, Average latency: 34.00 ms, Std: 1.47
Latency: 33.92 ms, Average latency: 34.00 ms, Std: 1.45
...

I'm having trouble understanding reference values in the docs. Are these transfer rates and latency as expected?

jakaskerl · Jun 4, 2024

Hi @AlanElkin
Yeah, I get the same.
Are you setting with dai.Device(pipeline, maxUsbSpeed=dai.UsbSpeed.SUPER_PLUS) as device:?

The above script you have sent has SUPER_PLUS enabled, I'm just rechecking because removing this results in similar FPS to what you have.

Thanks,
Jaka

AlanElkin · Jun 4, 2024

Hi jakaskerl

Again, thanks for your reply.

I'm using exactly the same code posted in the original post above, with the addition of pipeline.setXLinkChunkSize(0).

Just in case, this is the complete code:

import datetime
import depthai as dai

pipeline = dai.Pipeline()

# Configure camera
cam_rgb = pipeline.createColorCamera()
cam_rgb.setResolution(dai.ColorCameraProperties.SensorResolution.THE_12_MP)
cam_rgb.setFps(30)

# Configure XLinkOut (to send data from device to host)
xout_rgb = pipeline.createXLinkOut()
xout_rgb.setStreamName('rgb')

# Script node
script = pipeline.create(dai.node.Script)
script.setScript('''
    ctrl = CameraControl()
    ctrl.setCaptureStill(True)
    while True:
        node.io['out'].send(ctrl)
''')
# Connections: Linking camera to XLink input, so that the frames will be sent to host
script.outputs['out'].link(cam_rgb.inputControl)
cam_rgb.still.link(xout_rgb.input)

# Disable chunking for higher FPS (chunk size is for splitting device-sent XLink packets)
pipeline.setXLinkChunkSize(0)

with dai.Device(pipeline, maxUsbSpeed=dai.UsbSpeed.SUPER_PLUS) as device:
    queue_rgb = device.getOutputQueue('rgb', maxSize=60, blocking=False)

    tic = datetime.datetime.now()
    while device.isPipelineRunning():
        # Try to fetch data from queue. Returns either all data packets or None if there isn't any
        queue_frames = queue_rgb.tryGetAll()

        if len(queue_frames) > 0:
            toc = datetime.datetime.now()
            dt = (toc - tic)
            fps = len(queue_frames) / dt.total_seconds()
            print(f'Received {len(queue_frames)} frames in {str(dt)} --> Rate: {fps:.2f} fps')
            tic = toc

        # for queue_frame in queue_frames:
        #     msg = f'  [{queue_frame.getSequenceNum()}] '
        #     msg += f'Exposure: {queue_frame.getExposureTime().total_seconds()*1000:.3f}ms, '
        #     print(msg)

Is there anything else I can do to fix this? Or any other print you'd like to see?

AlanElkin · Jun 5, 2024

Just in case, this is my virtual environment, by running pip list:

Package                   Version
------------------------- --------------
AHRS                      0.3.1
anyio                     4.3.0
argon2-cffi               23.1.0
argon2-cffi-bindings      21.2.0
arrow                     1.3.0
asttokens                 2.4.1
async-lru                 2.0.4
attrs                     23.2.0
av                        12.0.0
Babel                     2.14.0
beautifulsoup4            4.12.3
bleach                    6.1.0
blobconverter             1.4.3
certifi                   2024.2.2
cffi                      1.16.0
charset-normalizer        2.0.12
comm                      0.2.2
contourpy                 1.2.1
cycler                    0.12.1
debugpy                   1.8.1
decorator                 5.1.1
defusedxml                0.7.1
Deprecated                1.2.14
depthai                   2.25.0.0
depthai-pipeline-graph    0.0.5
depthai-sdk               1.13.1
depthai-viewer            0.1.8
distinctipy               1.3.4
exceptiongroup            1.2.0
executing                 2.0.1
fastjsonschema            2.19.1
ffmpy3                    0.2.4
fonttools                 4.51.0
fqdn                      1.5.1
h11                       0.14.0
httpcore                  1.0.5
httpx                     0.27.0
idna                      3.6
ipykernel                 6.29.4
ipython                   8.23.0
isoduration               20.11.0
jedi                      0.19.1
Jinja2                    3.1.3
json5                     0.9.24
jsonpointer               2.4
jsonschema                4.21.1
jsonschema-specifications 2023.12.1
jupyter_client            8.6.1
jupyter_core              5.7.2
jupyter-events            0.10.0
jupyter-lsp               2.2.4
jupyter_server            2.13.0
jupyter_server_terminals  0.5.3
jupyterlab                4.1.5
jupyterlab_pygments       0.3.0
jupyterlab_server         2.25.4
kiwisolver                1.4.5
lz4                       4.3.3
MarkupSafe                2.1.5
marshmallow               3.17.0
matplotlib                3.9.0
matplotlib-inline         0.1.6
mcap                      1.1.1
mcap-ros1-support         0.0.8
mistune                   3.0.2
nbclient                  0.10.0
nbconvert                 7.16.3
nbformat                  5.10.3
nest-asyncio              1.6.0
notebook_shim             0.2.4
numpy                     1.26.4
opencv-contrib-python     4.5.5.62
overrides                 7.7.0
packaging                 24.0
pandas                    2.2.2
pandocfilters             1.5.1
parso                     0.8.3
pexpect                   4.9.0
pillow                    10.3.0
pip                       24.0
platformdirs              4.2.0
plotly                    5.22.0
prometheus_client         0.20.0
prompt-toolkit            3.0.43
psutil                    5.9.8
ptyprocess                0.7.0
pure-eval                 0.2.2
pyarrow                   10.0.1
pycparser                 2.22
Pygments                  2.17.2
pyparsing                 3.1.2
PyQt5                     5.15.5
PyQt5-Qt5                 5.15.2
PyQt5-sip                 12.13.0
python-dateutil           2.9.0.post0
python-json-logger        2.0.7
pytube                    15.0.0
PyTurboJPEG               1.6.4
pytz                      2024.1
pyusb                     1.2.1
PyYAML                    6.0.1
pyzmq                     25.1.2
Qt.py                     1.3.10
referencing               0.34.0
requests                  2.26.0
rfc3339-validator         0.1.4
rfc3986-validator         0.1.1
rosbags                   0.9.11
rpds-py                   0.18.0
ruamel.yaml               0.18.6
ruamel.yaml.clib          0.2.8
scipy                     1.13.0
Send2Trash                1.8.2
sentry-sdk                1.21.0
setuptools                69.0.3
six                       1.16.0
sniffio                   1.3.1
soupsieve                 2.5
stack-data                0.6.3
tenacity                  8.3.0
terminado                 0.18.1
tinycss2                  1.2.1
tomli                     2.0.1
tornado                   6.4
traitlets                 5.14.2
types-pyside2             5.15.2.1.7
types-python-dateutil     2.9.0.20240316
typing_extensions         4.10.0
tzdata                    2024.1
uri-template              1.3.0
urllib3                   1.26.18
wcwidth                   0.2.13
webcolors                 1.13
webencodings              0.5.1
websocket-client          1.7.0
wheel                     0.42.0
wrapt                     1.16.0
xmltodict                 0.13.0
zoomUtils                 0.0.0
zstandard                 0.22.0

erik · Jun 6, 2024

@AlanElkin still isn't meant for fast captures, you can use isp output for that, which should provide better latency/fps. But USB hubs do add additional overhead;

I get around 17fps when using OAK USB -> USB hub -> Mac
I get around 30fps when using OAK USB -> Mac directly

AlanElkin · Jun 10, 2024

Hi erik and jakaskerl,

Thanks for your suggestions. I've tried using isp instead but I still get slow fps, with no USB hub at all.

This is my test code (please check the first if-statement):

import datetime
import depthai as dai

pipeline = dai.Pipeline()

# Configure camera
cam_rgb = pipeline.createColorCamera()
cam_rgb.setResolution(dai.ColorCameraProperties.SensorResolution.THE_12_MP)
cam_rgb.setFps(30)

# Configure XLinkOut (to send data from device to host)
xout_rgb = pipeline.createXLinkOut()
xout_rgb.setStreamName('rgb')

if False:
    # Script node
    script = pipeline.create(dai.node.Script)
    script.setScript('''
        ctrl = CameraControl()
        ctrl.setCaptureStill(True)
        while True:
            node.io['out'].send(ctrl)
    ''')
    # Connections: Linking camera to XLink input, so that the frames will be sent to host
    script.outputs['out'].link(cam_rgb.inputControl)
    cam_rgb.still.link(xout_rgb.input)

else:
    cam_rgb.isp.link(xout_rgb.input)

# Disable chunking for higher FPS (chunk size is for splitting device-sent XLink packets)
pipeline.setXLinkChunkSize(0)

with dai.Device(pipeline, maxUsbSpeed=dai.UsbSpeed.SUPER_PLUS) as device:
    queue_rgb = device.getOutputQueue('rgb', maxSize=60, blocking=False)

    tic = datetime.datetime.now()
    while device.isPipelineRunning():
        # Try to fetch data from queue. Returns either all data packets or None if there isn't any
        queue_frames = queue_rgb.tryGetAll()

        if len(queue_frames) > 0:
            toc = datetime.datetime.now()
            dt = (toc - tic)
            fps = len(queue_frames) / dt.total_seconds()
            print(f'Received {len(queue_frames)} frames in {str(dt)} --> Rate: {fps:.2f} fps')
            tic = toc

This is the output:

Received 2 frames in 0:00:00.309474 --> Rate: 6.46 fps
Received 2 frames in 0:00:00.102032 --> Rate: 19.60 fps
Received 3 frames in 0:00:00.163704 --> Rate: 18.33 fps
Received 2 frames in 0:00:00.126305 --> Rate: 15.83 fps
Received 2 frames in 0:00:00.126683 --> Rate: 15.79 fps
Received 3 frames in 0:00:00.163115 --> Rate: 18.39 fps
Received 2 frames in 0:00:00.101160 --> Rate: 19.77 fps
Received 2 frames in 0:00:00.100358 --> Rate: 19.93 fps
Received 2 frames in 0:00:00.106135 --> Rate: 18.84 fps
Received 3 frames in 0:00:00.145875 --> Rate: 20.57 fps
Received 3 frames in 0:00:00.151619 --> Rate: 19.79 fps
Received 1 frames in 0:00:00.053159 --> Rate: 18.81 fps
Received 3 frames in 0:00:00.151166 --> Rate: 19.85 fps
Received 3 frames in 0:00:00.151327 --> Rate: 19.82 fps
Received 2 frames in 0:00:00.101480 --> Rate: 19.71 fps
Received 2 frames in 0:00:00.114627 --> Rate: 17.45 fps
...

As per Jaka's output, he receives Received 1 frames in every print, but I get not only 1 but a few as well. May that have anything to do with this issue?

erik · Jun 11, 2024

Hi @AlanElkin ,
Could you try some other host computer as well? I'd suspect the USB hub/card in the computer itself.

AlanElkin · Jun 11, 2024

Hi erik ,

I've tried running it on 2 other computers (so 3 hosts tested in total), all ubuntu based. To isolate any problem or difference coming from module/package versions, I've used luxonis' docker image to start a container, copied my test script and ran it on every computer. Unfortunately, I got same results: slow FPS.

Nonetheless, something curious is that, just by trying something different, when I used my hosts' USB type-C connector, it started receiving higher FPS, about 30 FPS!

So, I guess USB3 with type A connector isn't enough. In my case, I used Luxonis cable (type-A to type-C USB) + adapter. I'll try using a direct type-C to type-C USB cable to avoid using adapters at all.

Thanks for your assistance @erik and @jakaskerl !

erik · Jun 12, 2024

Hi @AlanElkin ,
Interesting! Does the USB C to C cable provide any better bandwidth / lower latency if you run the test scripts? My understanding was that it shouldn't change, as they\re both using the same USB3.1 interface.
Thanks, Erik

AlanElkin · Jun 28, 2024

erik

Sorry for the late reply, I took some days off.

Indeed, I believe the values don't change noticeably (except for the FPS). Here are the measurements:

With Luxonis cable (type-A to type-C USB):
- Around 18 FPS
- Average latency: 33.71 ms
- Downlink 2203.9 mbps
- Uplink 1903.9 mbps
With type-C to type-C USB cable:
- Around 30 FPS
- Average latency: 34.14 ms
- Downlink 2246.7 mbps
- Uplink 1931.1 mbps

jakaskerl · Jul 2, 2024

Hi @AlanElkin
Please redo the test with the updated bandwidth script. Pretty sure it was locking the transmission to 5Gbps (gen1) which caused them all to behave similarly.

Thanks,
Jaka

AlanElkin · Jul 5, 2024

Hi jakaskerl

With Luxonis cable (type-A to type-C USB):
- Downlink 2225.2 mbps
- Uplink 1912.1 mbps
With type-C to type-C USB cable:
- Downlink 3086.5 mbps
- Uplink 2689.2 mbps

Nice job fixing this bug!

Cheers