Friday, May 22, 2026

Q: I am leaning toward ReoLink's 12K cameras

A lot depends on your lighting. Reolinks, as a rule, are not great at night vision, mainly due to limited config options. I do like their POE TrackMix model though. Two lenses, so you get a zoomed and a wide stream, plus the wide stream helps with tracking in the zoomed view. Better IR night vision than many of their models too.

Do not bother with high MP cameras. Even 4K is often overkill, but 4MP generally looks much better than 3MP due to the sensors involved. Cameras are focused on a targeted range. Adding pixels generally does not let you zoom in much farther, though they might give you a wider view at the target range. That target focus range is not always specified in the specs for fixed-focus models, but it can often be inferred from the DORI numbers. Autofocus can help here, but be aware that a bug or bird might cause the camera to refocus, and it might not focus where you want it to. So, it's best to get it to focus on the target area and lock it there unless it is a tracking model. Never try to cover more area than what your cameras can see without moving.

For choosing models, the same rules always apply. Sort the distances to the target for each camera and decide how much detail you want to see. Match that to the DORI numbers in specs. Then factor in the width of the view angle to cover that target area. Lastly, factor in the lighting needed. Especially for color night vision, you want to compare min lux values and see real video. Many push the specs by reducing the exposures down to the point of ghosting moving objects. Or claim with internal white light on. IR night vision is not as bad. In both cases, you generally want to run without the internal lights. Accent lights for color, or IR floods for IR, are the way to go to keep bugs from visiting your cameras. Oh, and they should support RSTP or, preferably, ONVIF, which most of them do. Ignore the NDAA compliance BS. Do not assume ANY camera is safe, and block them from accessing the internet.

For NVR, go with a PC and Blue Iris or Frigate. Blue Iris has an easier interface and has an extendable AI, but is Windows-only and $100 for up to 128 cameras. Frigate is free and runs on Linux, but can be challenging to set up for non-IT types. Both work with cameras that support RSTP or ONVIF.

Tuesday, April 21, 2026

Updated copy/paste for what camera or system should I get.

While I love Unifi network gear, their cameras seem pricey and feature-limited.

For a doorbell, Reolink worked best for me (over Ring and Hikvision, which make most of the POE ones).

For other cameras, sort the specs to see the details you want in footage. Look for DORI numbers in the specs and ONVIF, or at least RTSP. Note that, as a rule, Reolinks do not have great night vision and offer limited image control. For color night vision, it gets tough. Many cameras, like Tapo, for example, crank down frame rates to brighten the image, which causes ghosting. Best is still Hikvision ColorVus, which will produce cloudy, day-like video if you can make out your hand in front of your face in the target area. Otherwise, if you can't see to walk without a flashlight or the moon, you want a camera that has IR for night vision.

Note: Avoid turning on in-camera lighting as it draws bugs and will white out close objects. Used IR floods or accent light as needed to add illumination.

Tracking is helpful, but it's best as an add-on. You want total area coverage recording 24/7 because no detection is perfect. Note: Reolink has a nice dual-lens tracking camera, but avoid the WiFi model.

NVR-wise, Frigate is free, and a lot of people seem to like it despite the setup pain. I tend to suggest Blue Iris for easy setup while still talking to most cameras and third-party integrations. I'd avoid all standalone NVRs. Especially Reolink's, which sets unknown passwords on the cameras, preventing direct access.

And of course, block all the cameras from the internet and the rest of your network. Preferably, only talking to the NVR PC.

Wednesday, August 6, 2025

Who is flying overhead?

Especially since the July 4th flooding. There have been a lot of helicopters in the sky over my neighborhood and a lot of posts on NextDoor and Facebook asking, "Who is up there?" So here is a quick guide to finding out.

Planes and Helicopters

Start off by going to https://www.flightradar24.com/2025-08-05/22:44/20x/30.58,-97.96/14.

Where:

2025-08-05 is the date you saw the aircraft in year-month-day format.

22:44 is the time you saw it in 24-hour format and the UTC timezone (add 12 hours for PM times and another 5 hours for Central Daylight Time, summer, or 6 hours during Central Standard Time, winter)

20x is the zoom. 20x is a good starting point.

30.58,-97.96 is the GPS to center the map on. (This is my neighborhood.)

14 is the zoom level. 14 is a good start for things overhead. If you are looking closer to the horizon, you will probably want to go to 11, which shows a wider area.

Click the play button to see the animation. It then turns into the pause button.

Once you see a likely aircraft, hit pause. Click on the craft's icon for info about it.

You can then click on "Route" to view the entire flight path.

Entering the tail number (R20045 in this example) in the search probably will return nothing because it is a military helicopter. So I'm using N3010X instead to show what a non-military aircraft would yield.

Clicking on that results gets you the history of its recent flights and basic info.

Sometimes adsbexchange.com has better info, though it is not as user-friendly. The equivalent URL for it would be.

https://globe.adsbexchange.com/?replay=2025-08-05-22:46&lat=30.430&lon=-97.933&zoom=11.0

Clicking on the military helicopter provides a bit more information.

Clicking on FULL DETAILS provides more information.

Drones

Drones are supposed to be using transponders now, too, but I would not count on them. You should see them on FlightAware, but I haven't seen any listed anytime I've looked. There are also apps that seem not to work well, according to reviews, and insanely expensive devices from Air Sentinel and DroneTag. Dedrone will not even disclose the price without contacting a salesperson, which usually means OMG pricing. Going DIY with an SDR dongle seems reasonable unless you are having issues with drones flying without IDs. Plus, that could give you a logging option that many of the others lack.

Monday, October 31, 2022

Another attempt at a quick paste response to "I want something with good night vision"

Best color night vision is Hikvision's ColorVu G2s. If you want IR night vision there are more options. I prefer Dahua or JideTech myself but many use the same larger sensors. For IR get the largest sensor you can. 1/1.8 (like Dahua IPC-HFW4631H-ZSA) of better if you can. Forget about stated LED ranges since you will want to turn those off anyway unless you like attaching bugs to your cams. Use seperate lighting like accent lights or IR floods mounted away from the cams.

Plus of course match the specs to the distance and view angle you need. But keep in mind most fixed focus cams are focused for 10-30 feet, depending on model, so for over 50 feet plan on something you can focus for the target range either motorized (usually zoom models only) or manually. Don't assume fixed model are easily adjusted as most are glued in place these days.

What ever you get make sure they support ONVIF or at least RTSP so you can mix brands as needed.

To dig deeper look at https://securitycam101.rmrr42.com/2020/05/getting-cam-specs-easy-way.html for info about how to use an online GUI tool tha lets you set up virtual cams on your site to get an idea of what diff models will be able to see.

Wednesday, September 28, 2022

DeepStack vs Sense AI and improving recognition.

First they both use:

YOLO models to compare found objects to.
The Sense AI API is a superset of the DeepStack API so something written for DeepStack should work with Sense AI too.
They use the same coding languages.

The main diffs between DeepStack and Sense AI are:

Sense AI is considered to be the more stable and supported (going forward) version. Though as of this writing you can probably find a lot more about working with DeepStack.
Sense AI will support more types of recognition and secondary operations like OCR.
But Sense AI is basically beta at this time so the real advantages are more subjective.

As explained in my DeepStack posts a lot of your success depends on:

Mode engine is run at. Medium is default. High will yield better results but at a cost of higher load. The reverse is true of Low.
Mode model was created for. The model ought to be made for the same or higher mode than the engine will be run at to see improved performance.
Trigger image might not be the best for recognition. You will want to tweak number and frequency of images sent to the AI when triggered to get better confidence in what is detected.
Same goes for minimum confidence percentages. Too low and you will get lots of wrong detections. Too high and you will miss things. Best to start low and see what object you want to know about are detected as what and at what confidence.
How close objects used in training match the ones in images used in detection. This is the biggie and why so many look to create their own custom models. Both for object types not in the default model and to more closely match expected targets in lighting and coloring. You can find tools for simplifying making your own custom models in my deepstack repo.

Friday, December 17, 2021

DeepStack training

My test set

My current test set can be downloaded from here.

Hardware

This workstation is set up for video editing so it is pretty fast. It was in it:

3.50 gigahertz AMD Ryzen 9 3950X 16-Core
64 GB of RAM
Samsung SSD 970 EVO Plus 1TB [Hard drive]
NVIDIA GeForce RTX 3080

Dataset prep

If you are looking to prep your own dataset you might want to look at the utils I have here.

You may want to look at Why Image Preprocessing and Augmentation Matter as well.

Method Compare

I used the original 91 pics with 194 labeled objects of 4 types of my test set:

Colab 2.973 hours
Jupyter hybrid method 18 minutes
local 0.299 hours (just under 18 minutes)

The models created appear to be the same. The latest model with the latest data set can be downloaded from here.

Local

Start here. Took me a while to sort what I was doing wrong as the instructions can be a bit misleading and python being at least 20 something on the list of languages I use the most. I finally figured it out while trying to debug the "local Colab" / jupyter method.

Be sure to install CUDA and CUDNN not just one. Remember the versions.

PyTorch

Instead of the suggested something like

pip3 install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio===0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Instead try

pip3 install torch torchvision==0.10.1+cu111 torchaudio torchtext -f https://download.pytorch.org/whl/torch_stable.html

The +cu111 bit seems to align to Cuda 11.1. Leaving off the other == bits lets it sort versions it needs itself.

Download training data to deepstack-trainer../deepstack/train

So assuming your workspace is C:\DeepStackWS

deepstack-trainer would cloned in C:\DeepStackWS\deepstack-trainer

and your training data would be in C:\DeepStackWS\deepstack\train

Then cd deepstack-trainer and run

python3 train.py --dataset-path --exist-ok "../deepstack"

Running this way not only avoids the notebook framework making automation easier but puts more progress info to the screen to see how things are going.

Options from train.py (with my notes)

--model, default='yolov5m', help='yolo model to use' (best option most cases)
--classes,default='', help='model.yaml path'
--dataset-path, required=True, help='path to dataset folder' (Should contain at least a train folder. May also contain test and validate folders. If folders exist they should contain some image files and copy of classes.txt or programs will crash.)
--hyp, default='data/hyp.scratch.yaml', help='hyperparameters path'
--epochs, default=300
--batch-size, default=16, help='total batch size for all GPUs' (>16 did not work for me)
--img-size, default=[640, 640], help='[train, test] image sizes'
--rect, help='rectangular training'
--resume, default=False, help='resume most recent training'
--nosave, help='only save final checkpoint'
--notest, help='only test final epoch'
--noautoanchor, help='disable autoanchor check'
--evolve, help='evolve hyperparameters'
--bucket, default='', help='gsutil bucket'
--cache-images, help='cache images for faster training'
--image-weights, help='use weighted image selection for training'
--device, default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu'
--multi-scale, help='vary img-size +/- 50%%'
--single-cls, help='train as single-class dataset'
--adam, help='use torch.optim.Adam() optimizer'
--sync-bn, help='use SyncBatchNorm, only available in DDP mode'
--local_rank, default=-1, help='DDP parameter, do not modify'
--log-imgs, default=16, help='number of images for W&B logging, max 100'
--workers, default=8, help='maximum number of data loader workers' (>8 did not work for me)
--project, default='name folder dataset folders are in', help='save to project/name'
--name, default='exp', help='save to project/name'
--exist-ok, help='overwrite if not first run instead of making new folder' (runTrain.bat assumes this to copy model to DeepStack instance)

Note increasing --batch-size or --workers caused and out of resources exception for me despite having plenty of unused RAM, GPU, CPU and HD. Though there maybe some missed option as Python only appears to be using 1 GB of RAM.

Gotchas

C:/w/b/windows/pytorch/aten/src/ATen/native/cuda/IndexKernel.cu:97: block: [0,0,0], thread: [95,0,0] Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"` failed.

This is caused by a bad mapping file with a class ID larger than the number of classes in the train\classes.txt file.

Note there is a similar problem using the LabelImg program. It will crash if it reads a map file with an ID larger than the number of classes in labelImg\data\predefined_classes.txt

Add debug

If you are seeing another kind of crash or odd behaviour with train.py try changing the set_logging method in deepstack-trainer/utils/general.py to look like this

def set_logging(rank=-1):

# logging.basicConfig(

# format="%(message)s",

# level=logging.INFO if rank in [-1, 0] else logging.WARN)

logging.basicConfig(

format="%(message)s",

level=logging.DEBUG)

Colab Method

The DeepStack docs mention doing this but they kind of assume you know what you are doing with Colab.

Open My edited copy of their notebook for cloud training. Notebook in this instance means project file not notes on how to do something which might be your first guess looking at it.

Click on copy to drive to get a copy you can edit.

I want to have my data and output persist between runs so I'm mounting my Google Drive for storage.

Go to / click in the mount drive section and press Shift-Enter to run it

Go to / click in the clone trainer section and press Shift-Enter to run it

Section 1

This takes a while. (Mine said 4 minutes.)

Note it loads old versions of packages so I was left with this error/result:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. torchtext 0.11.0 requires torch==1.10.0, but you have torch 1.7.0+cu110 which is incompatible. Successfully installed dataclasses-0.6 torch-1.7.0+cu110 torchaudio-0.7.0 torchvision-0.8.1+cu110

While this running you can work on preparing your data. See Preparing Your Dataset. If you just want to try it before creating your own data set you can find mine here. It is the one I'll building on to detect the wildlife types I have here.

Finally run the Add Your Dataset section which will take a fair bit of time. My set of 91 images took ~30 seconds per run and the default is 300 runs.

Note if you do not scroll the output often enough you will get a popup asking if you are still there. At which point I think it pauses your processing till you respond. So do not leave it to run overnight and expect to be done in the morning.

I finally got this result

Epoch gpu_mem box obj cls total targets img_size

299/299 9.26G 0.01827 0.0114 0.001359 0.03103 59 640: 100% 6/6 [00:30<00:00, 5.07s/it]

Class Images Targets P R mAP@.5 mAP@.5:.95: 100% 1/1 [00:00<00:00, 2.16it/s]

all 11 0 0 0 0 0

/usr/local/lib/python3.7/dist-packages/seaborn/matrix.py:198: RuntimeWarning: All-NaN slice encountered

vmin = np.nanmin(calc_data)

/usr/local/lib/python3.7/dist-packages/seaborn/matrix.py:203: RuntimeWarning: All-NaN slice encountered

vmax = np.nanmax(calc_data)

Exception in thread Thread-10:

Traceback (most recent call last):

File "/usr/lib/python3.7/threading.py", line 926, in _bootstrap_inner

self.run()

File "/usr/lib/python3.7/threading.py", line 870, in run

self._target(*self._args, **self._kwargs)

File "/content/deepstack-trainer/utils/plots.py", line 122, in plot_images

colors = color_list() # list of colors

File "/content/deepstack-trainer/utils/plots.py", line 32, in color_list

return [hex2rgb(h) for h in plt.rcParams['axes.prop_cycle'].by_key()['color']]

File "/content/deepstack-trainer/utils/plots.py", line 32, in <listcomp>

return [hex2rgb(h) for h in plt.rcParams['axes.prop_cycle'].by_key()['color']]

File "/content/deepstack-trainer/utils/plots.py", line 30, in hex2rgb

return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))

File "/content/deepstack-trainer/utils/plots.py", line 30, in <genexpr>

return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))

TypeError: int() can't convert non-string with explicit base

Exception in thread Thread-11: