Yes i know, but the time to transfert image through UART is too long, and my idea was to stock these images temporarely to this flash. I thought the vfs driver manage this page writing but you are right, there are some unstabilities. eg, sometimes all files i had on this flash disappeared, probably ...

Sorry i was not very explicit, because i proposed a solution and dit not with the idea. The idea is to have a template python function included in the github for newbie to understand how it works and test it. I think to do it by myseld, i should - create and add in py directory the files : py_file.c...

Well i dont understand what means "filesystem is cached in memory" because : F7 has : - sector (0) for boot of 32k - sector (1, 2, 3) for Fat of 32k each (32k x 3 = 96k) - sector 4 of 128k and sector (5 to 11) of 256k each for firmware each (128k + 7 x 256k = 1920K) H7 has : - sector (0) for boot of...

Is it possible to have a template lib for newbie like there are in imlib_config.h ?

It is not a function, it is the lens wich is fixed focus or auto focus !
If you use default lens of openmv, it is all fixed focus

Well thanks, but i dont know if i am sure to do a good and generic code well integrated enough to github it...
I tried to do a manual exposition and gain control (MEC and MGC) in function of a ROI luminance in a gray image.

What i know, you have algorythms to detect circle and rectangle which are the top and the side of a cylinder lol For image recognition, the pre-requesite is to have a sharpen image, and you can have this with - a fixed focus - global shutter - quick exposition - and sharpen algorythm Avoiding to to ...

Dear all, I have some problem of memory size with F7 cam if my python code is too big, not yet with H7, lol To improve this, and my knowledge, i would like to do and integrate my own python function with openmv github source. I have so two questions : - What is the simplest way to create my own pyth...

Dear all, When i connect a cam (H7/F7), there is a usb key mounted of 80Ko for F7 and 110Ko for H7 (without an SDCARD of course). I would like to know if it is possible to increase this size of this usb key because the overall size of flash is 2Mo (F7/H7) and the full firmware + bootloader is around...

Up for memory

I installed this : https://github.com/ntfreak/openocd
and
- it works for debugging bootloader (i did an frimware programmation with openmv IDE)
- but i have reset for firmware at led_init( hardfault ), and mp_init during mp_lookuptable(...) if i comment led...

Thanks, I will try this. What is your toolchain to debug H7 or F7 cam ? I would prefer to use the st toolchain, via stm32cubeMxProgrammer configurator and the Atollic True Studio IDE specially now where there is an AI cube package for cnn dedicated for CMSIS-NN. Unfortunately, i dont know how to joi...

Dear All, I try to do my own compilation with H7 cam, and it works, ouhou ! I did it with a virtuallbox i had with ubuntu and qtcreator than i use for F7 cam. For debugging with openocd (0.10.0), gdb-server and baremetal, i downloaded a stm32h7x.cfg here : https://github.com/ntfreak/openocd/blob/mas...

Dear All, I dont manage to get worked the cpufreq, here is the code : import sensor, image, time, cpufreq print(cpufreq.get_supported_frequencies()) cpufreq.set_frequency(200) #100 and 200 does not work print( cpufreq.get_current_frequencies() ) sensor.reset() sensor.set_pixformat(sensor.RGB565) sen...

Do you think is it possible to do a external ram shield ?

Have you have already think about how to add SDRAM for increasing heap, and size of main.py or cnn ?

Dear All, I want to take compressed snapshot (VGA/RGB565) with M7 every 3s to send it on UART at 115200bauds. To be sure to have best quality / size ratio, i triy to detect the best Q factor / file size ratio. Here is a code : import sensor, image, pyb, time, gc red_led=pyb.LED(1) red_led.off() sens...

Dear all, I try to analyse area of image, and i get different values of upper quartile for the same area. Here is the test code : import sensor, image, time sensor.reset() sensor.set_pixformat(sensor.GRAYSCALE) sensor.set_framesize(sensor.QVGA) sensor.skip_frames(time=1000) while(True): img = sensor...

Dear all, I try to analysesome areas of the image but i am not sure if i do it well because my results are not what i expected. For exemple, i created a bmp image, as i did it for masking, with the background at value 0x00 and an area in the middle with 0xFF value. I tried to do an AND boolean opera...

Thanks, I tried to modify the gain register at address 0x00 and read the db gain with sensor.get_gain_db(). Here is the code, and it is not linear between register and the db gain. import sensor, image, time sensor.reset() sensor.set_pixformat(sensor.GRAYSCALE) sensor.set_framesize(sensor.VGA) senso...

well, i see
- set_gainceiling : 2, 4, 8, 16 or 32 for x2, x4, x8, x16 or x32
- set_auto_gain : 2 to 32db

But i dont still understand teh relationship between gain and register ?
Thanks,

Dear all, I try to understand how works the gain in OV7725 camera and i am lost without speaking about analog gain. Firstly, there is the main gain at register address 0x00 which can be used with these both python functions where the gain is in db : - sensor.set_auto_gain( False, gain_db=(1 to 32db)...

Dear all, I try to understand how works the gain in the OV7725 camera, and i stopped on the gain ceiling. To set the gain ceiling there is 2 ways in python : - sensor.set_gainceiling(4) - sensor.set_auto_gain(True, gain_db_ceiling=4) which both write at the COM9 register but the c language codes do ...

Thanks, i effectively met this case.

How can i be sure of the lmean level before taking the first snapshot in RGB565 and VGA configuration ?

Dear all, I have different light conditions and i want to test the lmean before taking the first snapshot. Unfortunately, i cant do sensor.snapshot().histogram().get_statistics() in RGB565 and VGA configuration. Therefore i do it in smaller configuration i change after like this : import sensor, ima...

well, the uniform matrix is to reduce the number of lbp (256 possibility) to 59 most common.
So it is not usefull to change for specific use.
Thanks,

Well i understand than lbp_weights has a high value around eyes, medium value for mouth and ear, then low level for cheeks.
If i want to have a neutral use, i can set 1 for all the 7x7 matrix.

I dont understand what is for the uniform_tbl matrix ?

Thanks,
I red lbp.c and i discovered the weight are optimized for face recognition.
I would like to use for another use, what is the default values, and how can i define it to my own case ?

Wowwwww, i tested a 6x6 lbp on M7, and it works fine.
Which file from github i have to update to get this ?
Thanks,

Below is the simplified code, where you can define the number of line and column which define the number of areas where the lbp is defined and checked. I tested the M7 and H7 camera, and both get problem if i increase the number of line and column. With openmv 2.2.0 : For M7 (rev 3.4.1), in VGA : - ...

Dear All, I use lbp descriptor to recognize a pattern and, to filter local noise, i divide the image in several part with each a lbp descriptor. My problem is if i use more than 4 lbp pattern, whatever the resolution, i have frequently deconnexion and it does not seem stable. Do you have any knowled...

Nice, do you have some advice because i dont manage to quantize to openmv camera with : python2 ~/openmv/ml/cmsisnn/nn_quantizer.py --gpu --model ./quantizer.prototxt --weights ./snapshot_iter_240.caffemodel --save ./V22_Lenet.pkl I used a lenet network with my own image set. As nvidia digits is use...

Dear all,

Do you know if it is possible to use Nvidia digits software to train CNN and then convert it to openmv ?

Thanks,

Dear all, As everybody waiting the new H7 camera, I have M7 camera memory problem with cnn and heap size or heap variable allocation along time. If I would like to improve this, i am not sure to well understand how it works. Flash is divided like this - Bootloader : 32 Ko - Embedded Flash Drive : 96...

Ok i tested this and it seems ok sensor.__write_reg(0x24, 0x40) # New: 0x40 \ default: 0x75 / OPENMV: 0x58 sensor.__write_reg(0x25, 0x30) # New: 0x30 \ default: 0x63 / OPENMV: 0x48 sensor.__write_reg(0x26, 0xa1) # New: 0xa1 \ default: 0xD4 / OPENMV: 0xC3 sensor.__write_reg(0x0d, 0x71) # New: 0x71 \ ...

Dear, In place of turning auto gain and auto white balance off after 200ms, i changed the camera registers AEW and AEB in helloworld.py like this : import sensor, image, time sensor.reset() sensor.set_pixformat(sensor.RGB565) sensor.set_framesize(sensor.VGA) sensor.__write_reg(0x24, 0x40) #New: 0x40...

Dear all, I have got another camera with OV7725 and the image colors are more hot and more red compare to the M7 which the colors are more white and cold. This whiteness seems to blur the images like big aperture reduce depth of field. This other camera uses the OV529 ftom omnivision too, i dont kno...

lol, ok you mean the last current firmware.

I am not sure which function does this gamma correction.
Is it the function for histogram ?

Thanks,

Hi, the camera doesn't support gain/exposure for a small part of the image. As for the min and maxes, the gain has a max of 32 or so. The exposure isn't really limited by anything. But, if you set it very high the FPS will go to near 0. The exposure is based on the pixel readout line times which we...

Dear all, As i use a small part of the image, i try to control the exposure and gain to have the highest contrast with different ambiant lights. Is there a possibility to have the automatic gain and exposure control on a small part of the image ? sensor.set_windowing((128,128)) only cut the image (w...

To complete this post, i just discoverted than caffe have curve plotting python program. Here is the way i use it : 1- Create log file cd ~/openmv/ml/cmsisnn/ ./models/lettre_O_V3/fnt-train.sh >> ./models/lettre_O_V3/fnt-lettre_O_V3-output.log 2>&1 2- Draw plot Supported chart types: 0: Test_accurac...

yes you are right i jump to caffe forum
thanks,

Dear all, I try and understand how CNN training works by having the net image, learning curve and intermediate images of CNN training. For draw the net i use ~/caffe/python/draw_net.py ~/openmv/ml/cmsisnn/models/lettre_O_V3/fnt-lettre_O_V3_train_test.prototxt ~/openmv/ml/cmsisnn/models/fnt-lettre_O_...

Yes i scored 100% after train, test and quantize.
net.forward works well with an image with same size, ratio object/image size and binary characters.
I dont know how to get working the network with net.search at the moment...
Thanks,

Yes you are right, the explanations are very good and clear. To reduce the size of the network file, a simple way is to reduce the num_output in convolution layers of the train_test.prototxt architecture file. As you said, the cnn learn only as much as the datas have. My feeling is if you have a lot...

Hello, I trained and i try to quantize the network but it stops at "Network initialization done." step. The commands is : python2 nn_quantizer.py --gpu --model models/lettre_O_V1/lettre_O_train_test.prototxt --weights models/lettre_O_V1/lettre_O_iter_1300.caffemodel --save models/lettre_O_V1/lettre_...

well thanks, As you said, i suppressed directory but i still have a file.network around 90K what ever the number of letter and it is too big for the M7 ??? I try to train with different other parameters of solver.prototext to minimize the size without success at the moment. Any help would be appreci...

Hello all, I try to train a cnn with caffe to detect the letter "OK" or not. My first CNN is not very good and i would like to have your advices. I got fonts in png (white font on black background with 28x28px) for the letter O and K (real size is more around 120x120px, and the letter are more black...

Dear, I followed your log but my caffe was not in the same directory. Then i change path in train.sh and test.sh to get them working. My problem was simply because : - i did not compile caffe until the end (test, runtest and distribute) - i did not export pythonpath in the bashrc - i tried to use py...