RP2 DVP PIO driver consumes memory for some reason:

[sfe-SparkFro]

Replication code:

import cv2 as cv
from cv2_hardware_init import *
import gc

camera.open()
last_mem_free = gc.mem_free()

while True:
    mem_free = gc.mem_free()
    memory_used = last_mem_free - mem_free
    last_mem_free = mem_free
    print("Memory free: %d KiB" % (mem_free // 1024), end='\t')
    print("Memory consumed: %d KiB" % (memory_used // 1024), end='\n')

    key = cv.waitKey(1)
    if key != -1:
        break

camera.release()

Running after a soft reset prints:

Memory free: 7650 KiB   Memory consumed: 0 KiB
Memory free: 7611 KiB   Memory consumed: 38 KiB
Memory free: 7543 KiB   Memory consumed: 67 KiB
Memory free: 7476 KiB   Memory consumed: 67 KiB
Memory free: 7409 KiB   Memory consumed: 67 KiB
Memory free: 7343 KiB   Memory consumed: 65 KiB
Memory free: 7276 KiB   Memory consumed: 67 KiB
Memory free: 7208 KiB   Memory consumed: 67 KiB

Commenting out camera.open():

Memory free: 7649 KiB   Memory consumed: 0 KiB
Memory free: 7649 KiB   Memory consumed: 0 KiB
Memory free: 7649 KiB   Memory consumed: 0 KiB
Memory free: 7649 KiB   Memory consumed: 0 KiB
Memory free: 7649 KiB   Memory consumed: 0 KiB
Memory free: 7648 KiB   Memory consumed: 0 KiB

There shouldn't be any memory consumption here, because the driver is written to simply fill the buffer array, not allocate new arrays. So what would be causing the memory consumption?

[sfe-SparkFro]

Hmm...

import cv2 as cv
from cv2_hardware_init import *
import gc
import time

camera.open()
last_mem_free = gc.mem_free()

while True:
    mem_free = gc.mem_free()
    memory_used = last_mem_free - mem_free
    last_mem_free = mem_free
    print("Memory free: %d KiB" % (mem_free // 1024), end='\t')
    print("Memory consumed: %d KiB" % (memory_used // 1024), end='\n')

    key = cv.waitKey(1)
    if key != -1:
        break

print("Artificial delay...")
time.sleep(5)

while True:
    mem_free = gc.mem_free()
    memory_used = last_mem_free - mem_free
    last_mem_free = mem_free
    print("Memory free: %d KiB" % (mem_free // 1024), end='\t')
    print("Memory consumed: %d KiB" % (memory_used // 1024), end='\n')

    key = cv.waitKey(1)
    if key != -1:
        break

camera.release()

Memory free: 7588 KiB   Memory consumed: 0 KiB
Memory free: 7528 KiB   Memory consumed: 59 KiB
Memory free: 7443 KiB   Memory consumed: 85 KiB
Memory free: 7357 KiB   Memory consumed: 85 KiB
Memory free: 7273 KiB   Memory consumed: 84 KiB
Artificial delay...
Memory free: 7187 KiB   Memory consumed: 86 KiB
Memory free: 7173 KiB   Memory consumed: 13 KiB
Memory free: 7088 KiB   Memory consumed: 85 KiB
Memory free: 7001 KiB   Memory consumed: 86 KiB

I would have assume a big drop in gc.mem_free() after the delay, but apparently not. So the camera isn't consuming memory? But nothing else in the loop is consuming memory. So what's consuming memory?

[sfe-SparkFro]

Hmmmmm.....

>>> from cv2_hardware_init import *
>>> import gc
>>> gc.mem_free()
7771792
>>> gc.mem_free()
7771616                # Drop of 176
>>> gc.mem_free()
7771472                # Drop of 144
>>> gc.mem_free()
7771328                # Drop of 144
>>> gc.mem_free()
7771184                # Drop of 144
>>> camera.open()
>>> gc.mem_free()
7770656                # Drop of 528
>>> gc.mem_free()
7733456                # Drop of 37200
>>> gc.mem_free()
7695328                # Drop of 38128
>>> gc.mem_free()
7693744                # Drop of 1584
>>> gc.mem_free()
7656000                # Drop of 37744
>>> gc.mem_free()
7653200                # Drop of 2800

The driver appears to be not consuming memory as long as gc.mem_free() is not called, which is good and arguably not a problem. But it is a problem for the performance example, because it highlights how changes to the code affect memory consumption:

https://github.com/sparkfun/micropython-opencv/blob/08a7dbcbd5012fecafcfc14fe60e4e02da299505/examples/ex05_performance.py#L87-L95