Search results
-
CodeProject.AI Version 2.0
Hi, Could you attach GPU usage graphs from Task Manager, for both GPU 0 and GPU 1 (CPU's and the 970), of the benchmark, or BI load ? Thanks. -
CodeProject.AI Version 2.0
Possible sub-stream resolutions vary with the camera. E.g. a Dahua 5442 has 3 streams that is very nice. Can be configured e.g. to 1 full res., 1 full HD, 1 ~VGA. I love this approach. But for example a non-C Hikvision 2086 has only 720p 3rd stream, and so on. Dahua's full-HD third stream is... -
CodeProject.AI Version 2.0
It also depends on what number format it uses. NNs with multiple layers eat the binary precision. Also models must be re-quantized for each alternative format to give any result. So the TPU may be quick and low power on INT8 and FP16 but may not be able to supply the required precision. The... -
CodeProject.AI Version 2.0
Hi, This is how a startup looks like at my BI server. Originally the video decoding of both cameras was done on the built-in GPU (8th gen. i7, HD 630), GPU 0. Then the GPU 0 was busy doing 3D operations according to the task manager, with at about 30 % duration. CP AI was always set to GPU but... -
CodeProject.AI Version 2.0
In my setup it can use the CPUs GPU in DirectML if GPU is enabled. -
CodeProject.AI Version 2.0
Maybe some pages ago. I haven't tried them yet. -
CodeProject.AI Version 2.0
Don't know exactly but had similar issues. Because of so many moving objects like tree leaves and spider nets I set quasi-continuos inference by setting the trigger end times 3 seconds and multiple frames to AI. On the 'Alerts' tab there's a 'Minimum time between alerts' field. Have set the same... -
CodeProject.AI Version 2.0
They also use different models. Or even if their models are similar then the training differs I guess. So speed may be affected by their accuracy. I was wondering on the 'To cancel' list. It would be necessary to know what features can the models identify. Would like to try an ONNX analiser tool... -
CodeProject.AI Version 2.0
I think the .NET version can work with various GPUs because just tried a fresh CP.AI install on i7 6700 without BI present. It missed PIL, so 6.2 didn't work. But the .NET version worked, switched automatically to DirectML GPU mode and performed a 3 inference/second bechmark at 100 % HD 530 GPU... -
CodeProject.AI Version 2.0
I'm wondering on the .NET version that uses DirectML if compatible. -
CodeProject.AI Version 2.0
Thanks. Strange, that 'aiinput' is not amongst in my dropdown list. -
CodeProject.AI Version 2.0
That's a good point. I would be happier with 1080p. My cameras are external, looking at considerable area, where people can show up in various pixel sizes.. Do you think BI performs a resize to 640x640 for each source frame to the AI server? -
CodeProject.AI Version 2.0
Thank you. Will try that out. I've set the second stream bitrates extremely low for now to work with slow mobile networks as well. Will have to re-adjust them to normal range. Some of them are only VGA. But have some Dahua cameras that generate 1080p third stream that's very good for this... -
CodeProject.AI Version 2.0
Mine flats out around 24 inference/sec on pexels-huseyn-kamaladdin-667838.jpg . That's more than 12 times slower than yours. Could you disclose your setup, settings or that you think the most differentiating factor? I attached my BI settings. -
CodeProject.AI Version 2.0
I also had similar experience. ipcam combined classified more people in general. -
CodeProject.AI Version 2.0
WIll try that image a half day later. Your card is much better than mine (gtx 1660 super). But strange that yours running 10x faster. Will try with your image and we'll see how proportional it's.. Thanks. -
CodeProject.AI Version 2.0
Thanks, it's the same on my PC. .NET version with DirectML runs 1.5x faster than 6.2 with CUDA on the nvidia GPU. That's odd. The accucary of inferences is similar. They are obviously different as order of low-level operations, etc should exactly meet to have the same result, that I highly doubt... -
CodeProject.AI Version 2.0
Hi, The AI inference is done on the nvidia GPU. The procerssor's GPU decodes the video stream and the processor copies the frames I think, and possibly resizes. Unfortunately if the trigger frame is not an I-frame then it needs the deconding of multiple frames. That might be a reason why it's... -
CodeProject.AI Version 2.0
Hi, I got CP AI working eventually on GPU. Just stopped Face Detection and disabled half precision. It may work with face detection enabled, haven't tried afterwards. The AI evaluation loads the CPU significantly however. There are many trees and spider nets around the cameras (currently 24 are... -
CodeProject.AI Version 2.0
Yes. Yolo is a small net. It's also a question if BI resizes mainstreams to a certain resolution or not. I've read a post somewhere somebody wrote it resizes. I hope not. What I noticed is that if it calls the AI server with an 8 MP camera's mainstream images, for that I also set 10 additionals... -
CodeProject.AI Version 2.0
If the training algorithm is good then yes, a larger model should be more accurate. If the training algorithm limits then they may stuck. The model is an operation graph and set of operation filter weights. The operation graph is fixed, that's the YoloV5, etc... The filter weights are what must... -
CodeProject.AI Version 2.0
Great. Once I also want to update to some custom model to increase the accuracy. I think it's a lot of work. Model training might require a good training algorithm and thousands of annotated images.. -
CodeProject.AI Version 2.0
Have used the same model size. With deepstack I experienced similar accuracy between tiny and large models. Hence wanted to try CP AI. Don't know where they get the model from BTW. It would be reasonable if BI owns the models and just runs them on DS or CP servers. But I guess it's not the case... -
CodeProject.AI Version 2.0
Thank you for the plenty of information. I check it out in 1-2 days. Now the BI PC is away on the countryside and haven't set remote access up yet. -
CodeProject.AI Version 2.0
YoloV5 6.2 is for GPU. It uses the GPU, I checked that real time by an nvidia system monitor executable. It runs at bout 50 % peak and 15 % average GPU utilization and 2 % of 6 GB memory usage. As I remember other model returned % values for 'person', but the V5 6.2 on GPU always returns... -
CodeProject.AI Version 2.0
Thanks, I've disabled that already. Just seen your post a few days ago. But my idea is similar, somehow the execution results all zeroes, I think. But have no idea how, or if anything else causing this. -
CodeProject.AI Version 2.0
Hi, I'm just new to AI-based detection with BlueIris. Just tried CodeProject.AI server 2.1.9 with CUDA 12.1 on GTX 1660 Super executing Large models for 'person' detection. The settings are To confirm: person, To cancel: <field left empty>. It seems like running, from the server log with at... -
New Dahua Cameras will not remember NTP server
Well, not exactly. Just after I pressed the enter button and sent the reply it reverted back to the +1 hour error and the NVR enabled the DST again on the camera. So maybe the +1 hour timezone without DST on the NVR is what works in my case. These NVRs seem to be so overbearing. Don't know what... -
New Dahua Cameras will not remember NTP server
Keeps working. Thank you for pointing it out. -
New Dahua Cameras will not remember NTP server
Just trying your method. Now I start to understand :). So. Setting back the errorneous time to the correct one, like PC sync and disabling the NVR-set DST on the camera just keeping the right time already for several minutes. Let's see if it works. Thanks. I'm curious what happens after camera...