Replies: 36 comments 91 replies
-
|
Hi @paddywwoof. Nice to hear from you. You are doing a great job with with viewer. Feel freee to adobt the code. Or if you like you could join this repo, if you like to drive this viewer further as a separate project. I could take over the homeassistant integration. This is the only real requirement I have. |
Beta Was this translation helpful? Give feedback.
-
|
A first thinking about the components. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, sorry not to be more responsive. I got round to trying out your code the other day but not managed to get pictures to show yet - just a black screen I have to AltF4 out of (but no error messages). As a prerequisite for a general purpose replacement for the PictureFrame2020 in pi3d_demos I think people should be able to run it a) without installing the exif and mqtt dependencies b) on other platforms - I generally do all my development for pi3d on this ubuntu laptop but I try to make sure most things run on windows too. c) definitely RPi zero, 3 and 4. I don't know if that makes the location of config files and setup process just too complicated! (I also run repos such as pi3d accessed locally so I use the develop branch by doing import sys
sys.path.insert(1,"/home/patrick/python/pi3d")
import pi3dI was attempting to do this with your code for picframe) Your diagram makes sense as far as I understand the different boxes and arrows (I'm imagining structures such as instances of classes with arrows meaning methods at one end being called by called by methods at the other end) |
Beta Was this translation helpful? Give feedback.
-
|
I tried to run it in virtual box on ubuntu. But pi3d is crashing instantly in import sys worked for me, when I started to try things. The diagram is the general layout of the components. I will derive the class diagrams from it. |
Beta Was this translation helpful? Give feedback.
-
|
Would be nice to get it working on ubuntu with virtualbox. So I don't have to use my picture frame in the living room. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, thanks for trying that. It's ages since I used a VM but I think I had to install some additional dependencies. See http://pi3d.github.io/html/ReadMe.html#setup-on-desktop-and-laptop-machines the mesa library and pos numpy and PIL. It ought to work on Windows but you need the angle drivers in addition. Which I haven't updated for ages but should still work. I will try and do some more work on this this week. Paddy |
Beta Was this translation helpful? Give feedback.
-
|
Mesa is installed: I tried to run the demos. Minimal.py is working. Minimal_2d.py is craching with segmentation fault within an opengl lib. I will install the sources of pi3d an try to debug. Hope I can tell you where excactly it crashes. |
Beta Was this translation helpful? Give feedback.
-
|
Good that something works as that eliminates all kinds of things that need to be checked. but seg fault doesn't sound good. It might help narrow down what's causing it to try different demos and see if certain ones work and certain ones don't. I vaguely remember getting a seg fault with one of the demos some time in the past running under windows. Time to get my old windows laptop out and do some more tests. Paddy |
Beta Was this translation helpful? Give feedback.
-
|
If I add , use_glx=True to CreateDisplay, Minimal_2d_py is working, too. Don't know what this option is for. |
Beta Was this translation helpful? Give feedback.
-
|
Well done for narrowing that down so quickly. GLX is an alternative way of getting a display surface from the X11 windowing system. Which is used by most linux desktops UIs. On the raspberry pi it's possible to run the graphics without the X11 server (on zero, 2, 3) using a display surface via bcm_host (Broacom's display stuff) and EGL. So the likelyhood is that the seg fault is something to do with the EGL that's being used by the VM. This section https://github.com/tipam/pi3d/blob/master/pi3d/util/DisplayOpenGL.py#L89 is skipped if use_glx is True. Although seg faults should be rare in python the access to the OpenGL system and the GPU mean that there is lots of exposure to mapping variables and arrays directly to memory space. In python ctypes and C in general there are some rather vague things like I'm not sure what the disadvantage is of using GLX but it does allow making the X11 window partially transparent (provided there is a composition manager running such as compiz on ubuntu or xcompmgr on RPi) so the background alpha for the Display might need to be specified. Anyway, let me know if you get further info about what could be causing the problem. Paddy |
Beta Was this translation helpful? Give feedback.
-
|
Hi Paddy, I could nail it down to Setting use_glx=True and everythings works fine. Except the window is partially transparent as you mentioned above. To turn the display on the raspi on/of I'm using def display_is_on(self):
return True
cmd = ["vcgencmd", "display_power"]
state = str(subprocess.check_output(cmd))
if (state.find("display_power=1") != -1):
return True
else:
return FalseIf I always return true I can run and debug the picture_frame on ubuntu in the VM. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, that's great that you got it working (annoying that doesn't just work with the mesa EGL - probably the config isn't right, or the display is a 64bit pointer and should be 32. I've spent hours and hours trying to debug why the DisplayOpenGL doesn't work on different platforms) My temporary fix for I will spend some time finding my way round your code then try to make a list of ideas and suggestions. You will see on the pi3d_demos develop branch https://github.com/pi3d/pi3d_demos/commits/develop that there are a few mods added relating to the background darkening for the text and the ability to have two portrait images side by side. That latter probably needs to be thought through a bit more. Paddy PS I use VS Code on ubuntu - it runs fine. You really don't need windows for anything... |
Beta Was this translation helpful? Give feedback.
-
|
Helge, I'm thinking maybe the best way to approach this is for me to go through your code and change the minimum in order to put in the features that have been added to PictureFrame2020 since you diverged. During that process I will get more of a feeling for how the code is structured and I can probably make better suggestions for the way of handling the file_list etc. One issue I have come across straight away is that rotated images don't get picked up and corrected. Do you find your images are fine? Looking at PIL.ExifInfo I can only see one tag including 'rientation' which is >>> with open('/home/patrick/Pictures/tester/IMG_20201130_194139.jpg', 'rb') as f:
... tags = exifread.process_file(f, details=False)
...
>>> print(tags['Image Orientation'])
Rotated 90 CW
>>> print(tags['Image Orientation'].values)
[6]
>>> print(tags['Image Orientation'].tag)
274However I don't want to alter the existing code if it works OK for your images. PS also I'm not sure I understand why get_exif(key) returns {key: val} rather than just val - and why val is the printable form rather than the whole object (so the receiver can do the unscrambling) # in exif2dict ####
def get_orientation(self):
val = self.__get_if_exist('Image Orientation')
if val is not None:
return int(val.values[0])
else:
return 1
# then in model ##
try:
exifs = exif2dict.Exif2Dict(file_path_name)
orientation = exifs.get_orientation()
for exif in self.get_model_config()['image_attr']:which seems to rotate my images correctly |
Beta Was this translation helpful? Give feedback.
-
|
Hi Paddy, you are right. I used the wrong tag. "Image Orientation" is correct. I checked al my images and none of them have any orientation tag. I alwas shoot in raw. So in post processing my images have always the correct orientation. Only the jpgs of my mobile phone have the tag set. But none of them is on my picture frame. My exif2dict class is recycled. I used it get a key:value tags to send them over mqtt to my home assistant to display the image meta data. |
Beta Was this translation helpful? Give feedback.
-
|
Great, I will try to change as little as possible to get the new functionality working and hopefully not break anything! This is the problem with a 'general' picture frame: nobody uses all the features but somebody always uses something that somehow never got tested!! |
Beta Was this translation helpful? Give feedback.
-
|
I've put my initial mods in a I will look at the OSM description lookup code next. As a first step I've made the image info into instances of class Pic so the properties can be accessed in a more friendly way (a dict could have been used), part of the reason for this (apart from readability) is to allow the info to be passed and altered by the PS at the moment |
Beta Was this translation helpful? Give feedback.
-
|
This is my first python project and I'm an absolute beginner to this language. I usually code in C++. I would provide all info to the viewer in a structure (dict). I would keep all database, or file_list manupulation inside of the model. This might be a little overhead. But hiding the internal structure allows you to refactor the code later, without side effects. This is also the reaseo, why I tried to avoid global variables. |
Beta Was this translation helpful? Give feedback.
-
|
I haven't cacked your code yet. But maybe it is fine to pass the Pic object so you can use its functions inside of the viewer. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, yes all those double under-scores were an indication that you were trying to make python conform to the disciplines of 'proper' structured language! Global variables are especially tricky in python as the interpreter will try to figure out what scope you intended your variable to belong to. There is a system, but it often trips people up. It doesn't seem to be used so much in python but the nearest it has to a |
Beta Was this translation helpful? Give feedback.
-
|
The problem might be that some functionality done in |
Beta Was this translation helpful? Give feedback.
-
|
Good point. If you see exif2dict as a genral class to retrieve image meta data for the model, it should have have a function to return the size. To turn the display on/of on linux I could use (xset dpms force on). But this requires glx = true. So should we create the display on raspbery with glx=false and for all other systems with glx=true? I have no idea how to handle this on windows (maybe, just show a black image) but this wouldn't activate the power management of the display. |
Beta Was this translation helpful? Give feedback.
-
|
some thoughts on db: While looking at the OSM location system in PictureFrame2020 I started thinking about the pros and cons of storage, at the moment the location is just stored in a CSV type file, so I looked at sqlite3 again. For the simple lookup it's fine but when I thought about the image list (__file_list) I could see some disadvantages to sqlite:
A solution would be if the db select creates a python list of ids that can be traversed (including backwards) and the info read for display (or written to the database as each image has the exif info extracted). For a CSV (or JSON) system then it would also make sense to hold the full list of image info as a dict and have the selection held as a list in a very similar way to the db mechanism. On startup all the image files would still need to be read from the SD dictionary structure (as well as checking from time to time for additional files) but in order to allow selection according to criteria the exif would be needed. The reason that the delayed exif reading was introduced was because for some users it was taking so long to read through all the files (>20minutes). So maybe a compromise would be for the exif reading to take place in a second thread which would add records to the db (or CSV and dict) in the background. The controller would need to keep informed and be able to handle additional records being added - maybe do a reselection when it detects a change. I suppose, to start with, the db functionality could be wrapped in its own class with 'select','update' and 'insert' methods and delay deciding how those are implemented? Paddy |
Beta Was this translation helpful? Give feedback.
-
|
I had similar thoughts.
|
Beta Was this translation helpful? Give feedback.
-
|
Good (that you thought similar (as did Jeff on the pi3d_demos thread)) The main issue I can see is that the first time through, or after a load of images have been added (just back from holiday, if those will ever happen again), the meta info is missing so following your outline: The full list would be traversed and the meta data obtained - the only problem with lazy load filtering would then be if the selection was too tight, in which case it might take a long time for nothing to show up. The alternative to lazy loading for each picture just before showing would be the background threaded loading, however there would be an issue with that in that if it was also getting the nominatim OSM location description as they give an absolute maximum query rate of one per second - and prefer lower. - And too tight filtering would be the same either way, no way to expedite that. So, to summarize, the traditional db functionality isn't really going to be used in the picture frame, as we envisage reading single records from it one at a time and checking the extracted content against the specific criteria. I suppose having the info in a db might be useful elsewhere for people to do some kind of analysis of where and when pictures had been taken - but I imagine all that has already been done. |
Beta Was this translation helpful? Give feedback.
-
|
Good point. But if you want to filter, you will need the content to befiltered. So starting with scanning just the file list (ids) will provide you an unflitered list, where you can start to show images. Maybe you start lazy load with the actual selected subdirectory. We could show an indicator, that indexing the files for the current selected directory is not done and filtering might not work correctly. I would expect, that indexing happens not so often. If I add my new 400 vacation images. This scan should be pretty fast. But I wouldn't care if I have to wait a couple of minutes to apply the first filter (except supdirectory). |
Beta Was this translation helpful? Give feedback.
-
|
I've made another upload to my fork paddywwoof@3c9c161 it has basic simple caching (not yet sqlite) I think the only functionality still to add is the paired portrait idea. Paddy |
Beta Was this translation helpful? Give feedback.
-
|
Another commit paddywwoof@4aab48e . "Portrait pairs" system now in. I'm not sure this is a nice way of doing it: The viewer gets passed a tuple of either (pic, None) or (pic1, pic2) if they are to be shown as pairs. As discussed above this functionality (and others) only works properly once images have been "processed", at the moment this happens as each picture is being got by Also this is just writing cache info to txt file, not yet looked at sqlite. And I haven't put in the small bit of code for heif decoding. PS patched a few errors thrown up running with random 'real' files on the 'real' picture frame in my kitchen. The whole exif extraction might be better inside a try block as there seem some odd formed results returned by exifread!! I also need to check out the MQTT which I haven't touched or checked in any way. |
Beta Was this translation helpful? Give feedback.
-
|
I pulled your commit. To get it working, I had to make some adjustments: paddywwoof/picture_frame@develop...helgeerbe:ubuntu |
Beta Was this translation helpful? Give feedback.
-
|
Sorry, I had a config file in the same directory I was running picture_frame.py from and didn't post that to github - though I see you patched it from the default in model.py. paddywwoof@073e538 How long does it take you to find an address? I'm assuming you mean geo_reverse.py. I did this >>> import time
>>> from picframe.geo_reverse import GeoReverse
>>> geo = GeoReverse('[email protected]', 'junk.txt')
>>> def timer():
... start = time.time()
... addr = geo.get_address(547123,-17987) #i.e. 54.7123, -1.7987
... print(time.time() - start)
...
>>> timer()
0.13068056106567383and generally I've been finding the lookup to be measured in milliseconds rather than tens of seconds. Are you using your proper email address for the geo key? Does the text appear fine if you use PS you can see that I have used integer values for the lat and lon, that's because I envisage them being indexed fields in a database or being matched against and floats are always tricky to 'equal' to something. And slower as db fields. |
Beta Was this translation helpful? Give feedback.
-
|
Today
|
Beta Was this translation helpful? Give feedback.
-
|
I foud that you are using uv_flat. I don't have this shader. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, that sounds good. I've just pushed a couple of changes to my repo. I made get_next_file bi directional. Going backwards it doesn't need to return the Pic object but that should be an inexpensive process as it should always be iterating over records that have already had their exif data extracted. It now correctly skips over images with dates outside the selection criteria going forwards and backwards. If the image is in a portrait pair then the right hand image isn't filtered, but otherwise seems good. I'm increasingly leaning towards a system: On a different topic, I've had a go with my RPi4 and I can't get it to run without blend_type: "blend" # default="blend", choices={"blend":0$
font_file: "./data/fonts/NotoSans-Regular.ttf"
shader: "./data/shaders/blend_new"
show_text_fm: "%b %d, %Y" # default "%b %d, %Y", format to show$and in picture_frame.service [Service]
Type=idle
User=root
WorkingDirectory=/home/pi/.local/picture_frame
ExecStart=xinit /usr/bin/python3 /home/pi/.local/picture_frame/picture_frame.py configuration.yamlI haven't installed picture_frame properly just created ./local/picture_frame and copied picframe/ and data/ into it, as well as picture_frame.py and configuration.yaml but I think something like this will work for your proper setup. Paddy
|
Beta Was this translation helpful? Give feedback.
-
|
Finally it seems to work now: All elements were generated automatically in homeassistant. I haven't tested everything. But it seems to work in general. So mqtt is now optional and I put it in its own class In mqtt_config = m.get_mqtt_config()
mqtt = None
if mqtt_config['use_mqtt'] == True:
from picframe import interface_mqtt
mqtt = interface_mqtt.InterfaceMQTT(c, mqtt_config)
mqtt.start()Your MQTT is running in a separate thread. Is the pi3d.keyboard interface also running in a separate thread, or do I have to put it in its own thread? |
Beta Was this translation helpful? Give feedback.
-
|
Helge, controller and interface_mqtt look much tidier now. I will have a look at the controller state callback... but if it works maybe don't worry how pythonic it is! The homeassistant interface is fantastic! I'm not sure about the use of pi3d.Keyboard uses different systems depending on the system it's running on - but all run the key capture and buffering in their own process. On X (i.e. ubuntu and RPi4) it uses the Xserver, on the RPi without X running it uses curses and on windows it uses pygame. But there is no background loop you can hook functions into to make the key presses proactive - normally there is a read done each display loop which returns whatever key was pressed. To put it into its own interface you would need a basic threaded loop to do the checking then fire off a control function accordingly. I will merge all your modifications over and get it all working. Then contemplate getting all the sqlite stuff in there! PS had a look at the state callback and, though it seems a bit of a 'hidden side-effect' with the function being reset by a remote object, I don't think you can really get round it. The alternative would possibly to have an instance of the interface_mqtt set in the Controller instance but you would then have to test if it had been set. You could also do the setting in a function to make it clearer (for when we look at this in a year's time!) But having circular references like this is a bit of a minefield. ...
self.interface_mqtt = None
def set_interface_mqtt(self, interface_mqtt):
self.interface_mqtt = interface_mqtt
...
pics = self.__model.get_next_file(self.date_from, self.date_to)
if self.interface_mqtt is not None:
self.interface_mqtt.publish_state(pics[0].fname, pics[0].image_attr)but if the existing system works I would think it's fine as it is. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, I pushed the first draft of a version using sqlite3 on my repo. It's got 38 TODOs!
|
Beta Was this translation helpful? Give feedback.
-
|
I spent a while trying to get RPi4 working with In the end I had to go back to the system of changing directory to I don't seem to be getting any exif data from heic files (and I had to catch an error from exifreader generated by heif files) they display OK. Also I added in a first time flag for the cache checking. It's been running for a while now on two RPi so hopefully mainly fixed... |
Beta Was this translation helpful? Give feedback.
-
|
Helge, I think I might have found a way round this. It's possible to do pic_dir = "~pi/Pictures"
pic_dir_path = os.path.expanduser(pic_dir)
# gives /home/pi/Pictures even when running as rootThe other thing Manuel asked: I've not installed any of this properly but in normal use is there any reason why the picture_frame.py as well as configuration.yaml (and picureframe.db) couldn't live in a normal 'visible' directory so they could be edited, or viewed remotely, more easily. That's the way I've been testing it but is there a issue with doing it that way? Paddy |
Beta Was this translation helpful? Give feedback.
-
|
Paddy, there is no other reason that setuptools is using ./local to store date. I simply don't know how to change it. I have read, that there is an alternative for setuptool. ~pi works. But if you install under a different user or other system you have to change it anyway. So if there is no other chance, you could use ~p or the absolute path. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, You're right if the configuration.yaml needs to be edited then why not just put in absolute paths on the RPi4. So I'm not sure which way is best. Sticking to the default location for setuptools is probably the target. With pi3d, whenever I've installed it with pip, it gets put in When you install picture_frame (the proper version prior to this hacking) could you put picture_frame.py and configuration.yaml in a different folder and run them OK (i.e. would the |
Beta Was this translation helpful? Give feedback.
-
|
Some more thoughts: controller holds a dict of where_clause which are changed or updated by the functions, something like {'date_to': 'exif_datetime < 1610789629', 'rating': 'rating > 4'}then ' AND '.join(where_clause.values()) and a similar sort_clause. For sorting, rather than have the list shuffling is a special function we could try to move it into the image_cache by having a The other thing might be to have a timer file that worked in a similar way to the mqtt in that it had a copy of the controller instance and turned the display on and off or the brightness up and down etc. This might or might not be nicer than having crontab and mqtt external systems. |
Beta Was this translation helpful? Give feedback.
-
|
Switched to more general system of filtering where the SQL is generated in the controller see paddywwoof@6285dfc |
Beta Was this translation helpful? Give feedback.
-
|
Added an MQTT picframe/stop and tidied up the kenburns system. Also reverted the ~pi/ back to ~/ as discussed above. We probably need some clear instruction on how to download for testing without installing to enable people to try it out. |
Beta Was this translation helpful? Give feedback.
-
|
I hope I have time to pull your changes and test it. Impressive work Paddy. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, that's cool - and very straightforward. However I can't for the life of me get that to work on the RPi4 where it all has to be passed as command line argument to All platforms seem to work OK if I move picture_frame.py up a directory. That way from picframe import model, viewer_display, controllerworks, as picframe is a subdirectory and is then interpreted as a module. As I think @sapnho and @jgodfrey (and you) are all using RPi3 for testing it should be fine to run as module initially. For the time being I will hold off making any other alterations. I would find it handy to have a simple httpserver running but that's not much code if it does nothing more than relay values to Controller (and doesn't have any fancy HTTP or javascript!) And I will only add that once people have dug out all the initial bugs! |
Beta Was this translation helpful? Give feedback.
-
|
Paddy, we could put something like this as a shell script in the root directory: #!/usr/bin/python3
# -*- coding: utf-8 -*-
import re
import sys
from picframe.picture_frame import main
if __name__ == '__main__':
sys.argv[0] = re.sub(r'(-script\.pyw?|\.exe)?$', '', sys.argv[0])
sys.exit(main())setuptools is creating this in .local/bin. If this is in $Path you can run it from anywhere |
Beta Was this translation helpful? Give feedback.
-
|
I played with an keykoard interface """Keyboard interface of picture_frame."""
import logging
from pynput import keyboard
from picframe import __version__
class InterfaceKbd:
"""Keyboard interface of picture_frame.
This interface interacts via keyboard with the user to steer the image display.
Attributes
----------
controller : Controler
Controller for picture_frame
Methods
-------
"""
def __init__(self, controller):
self.__logger = logging.getLogger("interface_kbd.InterfaceKbd")
self.__logger.info('creating an instance of InterfaceKbd')
self.__controller = controller
listener = keyboard.Listener(
on_press=self.on_press,
on_release=self.on_release)
listener.start()
def on_press(self, key):
try:
print('alphanumeric key {0} pressed'.format(
key.char))
except AttributeError:
print('special key {0} pressed'.format(
key))
def on_release(self, key):
print('{0} released'.format(
key))
if key == keyboard.Key.esc:
self.__controller.stop()
# Stop listener
return FalseThis works fine. But requires X. Would you propose to use pi3d.Keyboard? This is non blocking, right? So I have to put it in a separate thread and polling for a key press (with sleep(1))? Or do you have a better idea? |
Beta Was this translation helpful? Give feedback.
-
|
Helge, yes it is a problem. I think most people run the picture frame without the x server - though on the RPi4 you have to have it there. You could use a pi3d.Keyboard (which will use X if it's available, otherwise it uses ncurses) The problem is that the InterfaceKbd instance has to be created after the pi3d.Display instance - as it uses that for the central storage of keyboard and mouse info (when X is available). Running in it's own thread seems to be OK, however there is obviously a mess of references as this gives a seg fault quite often on my laptop - when it closes down, runs OK until then. import sys
import threading
import time
sys.path.insert(1,'/home/patrick/python/pi3d')
import pi3d
from picframe import __version__
class InterfaceKbd:
def __init__(self, controller):
self.__controller = controller
self.__keyboard = pi3d.Keyboard()
self.__keep_looping = True
t = threading.Thread(target=self.__loop)
t.start()
def __loop(self):
while self.__keep_looping:
key = self.__keyboard.read()
if key == 27:
self.__keep_looping = False
elif key == ord('a'):
self.__controller.back()
elif key == ord('d'):
self.__controller.next()
time.sleep(0.025)
self.__keyboard.close() # contains references to Display instance
self.__controller.stop()and in picture_frame.py ...
c.start()
kbd = keyboard.InterfaceKbd(c)
c.loop()
... |
Beta Was this translation helpful? Give feedback.
-
|
Under Ubuntu with X it works like a charm. Cpu load is slightly higher, but this seems to be OK. I don't see any errors. I will test it tomorrow on raspberry without X. |
Beta Was this translation helpful? Give feedback.
-
|
On rasbian lite without X, everything works fine. No seq faults. What is your setup to get the seg faults? |
Beta Was this translation helpful? Give feedback.
-
|
That sounds good news. Maybe it's something to do with the lib mesa-utils-extra - there is very little memory leakage control in C and the OpenGL is all ctypes calls. Horrible really! I'm using standard ubuntu 20 LTS on an old dell i7 laptop. I will stick some "print('here')" debugs in around the place to narrow down at what point the seg fault happens. |
Beta Was this translation helpful? Give feedback.
-
|
Paddy, I'm still struggling with the setup. If you provide a binary package all is installed under the .local directory.
from the site-packacke. So no more ./local/picframe. What do you think? |
Beta Was this translation helpful? Give feedback.
-
|
Helge, I'm all for simple installation so if you can make that |
Beta Was this translation helpful? Give feedback.
-
|
Paddy,
To get it working, I had to move the folders into the picframe folder to become part of the site-packages. To test it. You can run |
Beta Was this translation helpful? Give feedback.
-
|
Helge, that looks really neat - I will try it out (probably tomorrow morning now) Did you try the htp server? I found a minor bug which I fixed here Then it just needs a bit of tidying up and some documentation! |
Beta Was this translation helpful? Give feedback.
-
|
I just started the webserver. Good job. Tomorrow I will test it in detail. It took me a while to find out, that I missed entering index.html in the url. Normally, this ist the default. So missing is a publish handler to get details, of the actual image. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, there are a few things missing (also see this TODO on index.html ;) ) When I mentioned it to my daughter the first thing she asked was "is there a drop-down list of directories" - obviously there could be but the whole thing quickly becomes much more complicated (so I said no!). However I think a |
Beta Was this translation helpful? Give feedback.
-
|
On a different but related issue. I have just pushed up a revised version to paddywwoof/picture_frame develop branch. c.start()
server = interface_http.InterfaceHttp(c, "/home/pi/picture_frame/html")
...Fix for my muddle between using 'folder' and 'directory' paddywwoof@5588969 I still get seg faults running on this laptop. I've not tracked down exactly where but I'm pretty sure they're due to accessing after freeing, possibly due to the various thread looping. It might only happen when the processor is fast enough. I will carry on putting Going to try the new set up on as many RPis as I can now. Paddy |
Beta Was this translation helpful? Give feedback.
-
Nice. I'll need to study this as I really know very little about |
Beta Was this translation helpful? Give feedback.
-
|
Jeff, it occurred to me later that I forgotten to convert the RGB vectors to floats. Leaving them as uint8 generates essentially random numbers from overlapping and wrapping! Fixed above. c_max, c_min = centroids.max(axis=1), centroids.min(axis=1) # max, min for each centroid
c_lum = 0.5 * (c_max + c_min)
c_sat = (c_max - c_min) / (255.0 - np.abs(c_lum * 2.0 - 255.0)) # should check for lum == 255
ix_order = np.argsort(c_sat)[::-1] # indices to sorted values - reversed
print(centroids[ix_order].astype(np.uint8))
print("frequency", counts[ix_order])
print("luminosity", c_lum[ix_order])
print("saturation", c_sat[ix_order]) |
Beta Was this translation helpful? Give feedback.
-
Done! interface-http builds its own path for image download. "pic_frame" plus actual selected "subdirectory". If image is not found there, the "no_img_file" is returned. This nails down the image download just to the configured directories. |
Beta Was this translation helpful? Give feedback.
-
|
Helge, that's neat. I've merged it into my repo but changed it slightly. If it needs to be done the way you set it up I'm happy to revert, however I think it is simpler to make the new variables part of the HTTPServer rather than having to run init on the HTTPRequestHandler. We were already doing that with controller so then we had two references: I'm not sure how you want the image rendering to work and it might be fine as it is but I can't figure out how to show most of my images. I suspect my setup will be similar to most: So if I set the subdirectory to paintings I get all the paintings in random order even though there are no images. I would expect to be able to be able to render the image actually showing on the picture frame but for that to work I would have to be able to enter |
Beta Was this translation helpful? Give feedback.
-
|
Helge I've added a tiny bit more code to enable the http interface to get the current image, whatever that happens to be. paddywwoof@901c959 don't know if it's any use for your setup but it might be useful for some Paddy |
Beta Was this translation helpful? Give feedback.
-
|
Paddy, you are absolutely right. I don't know what I had in mind. I just need the image name to alter the url (not, tu run in any caching). Using the path, as you did, from controller. That will work. I wasn't aware that calling super provide access to class variables of server. So I'm totally fine with your changes. I'm struggling with the webserver. Calling from the browser will load the image. Calling from homeassistant within a markdown document |
Beta Was this translation helpful? Give feedback.
-
|
Hi that's tricky. The only thing I can suggest is putting |
Beta Was this translation helpful? Give feedback.
-
|
That' what I get in the output |
Beta Was this translation helpful? Give feedback.
-
|
OK, I have an idea. I'm running homeassistant ssl encrypted. I debuged server.py. The request was unreadable. If I turn off ssl. It works and the image is displayed. I assume, that homeassistant doesn't care about http or https. So I try to enable ssl with http.server. |
Beta Was this translation helpful? Give feedback.
-
|
Ah, presumably the interface to homeassistant is exposed to the internet which makes the interface_http a bit of a loophole. You might be able to do something like https://blog.anvileight.com/posts/simple-python-http-server/ which also makes me think we should use I think the basic server is still not very secure for other reasons so shouldn't be allowed access to the outside internet. |
Beta Was this translation helpful? Give feedback.
-
|
OK, ssl is working and the image shows up. This took me the whole evening. I don't expose any server to the internet. But I tried to get rid off of the 'unsecure' messages, when working in the browser. So all my servers are using letsencrypt certificates. I'm happy that it works now. This makes the setup for homeassistant much easier. |
Beta Was this translation helpful? Give feedback.
-
|
Is it possible for people to use ssl without installing anything else (i.e. just use the built-in ssl module) and some kind of use_default_certs? In this instance it's probably a good thing to have the 'unsecure' message as it will still be unsecure even with ssl encryption!! |
Beta Was this translation helpful? Give feedback.
-
|
One can use selfsigned certificates. But you have to import them to the browser as trustable. That's why I use letsencrypt. The root certificates are already known to the browser. But this ended for me in an other project. I had to write my own certbot plugin. I think normal user wouldn't use ssl. Only nerds like me. But its good to know its working. To use its externally one should use a vpn, or put everything behind a proxy. Never expose your hosts to the internet directly. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, I do, but I will start with the Pi3 tomorrow, updating my main blog post as I go along, and probably asking you guys a few questions. I lost track of which thread is the right one, maybe I'll just open a new thread with questions only relating to the documentation. Once that works, happy to test the Pi3 which should be quick. Which repository is the one to use? Helge or Paddy? Not sure if there is a difference. |
Beta Was this translation helpful? Give feedback.
-
|
Wolfang, use my repro and open a new Q&A discussion. I will put all my answers in the wiki https://github.com/helgeerbe/picframe/wiki and post the link as an answer. Goal is, to have the wiki ready, before release. |
Beta Was this translation helpful? Give feedback.
-
|
Wolfgang can you try this on a pi4. https://github.com/helgeerbe/picframe/wiki/Running%20As%20Service |
Beta Was this translation helpful? Give feedback.
-
|
I've now switched to using helgeerbe/picframe dev branch as Helge has given me admin rights. (Hope that's what you intended) |
Beta Was this translation helpful? Give feedback.
-
Was going to ask how we wanted to handle this moving forward. I'll do the same. |
Beta Was this translation helpful? Give feedback.
-
|
Perfect. But create feature branches, when working on issues that take more than 1 commit. So dev should always be working. |
Beta Was this translation helpful? Give feedback.
-
Moved from Issues. Original post from @paddywwoof
Hi @helgeerbe you've done some great stuff here. I've started to think about refactoring PictureFrame2020 (not just because of the bad 2020 karma) I had intended to break it into different functional classes very much along the lines you have but I'm not sure whether to just adopt your code as it is and add in the 'improvements' which have happened since you diverged. If I can keep it compatible with Home Assistant but still work without it - the main thing is that Wolfgang's setup will work (almost) unaltered; there's a lot of really good work in his picture frame blog and the less he has to re-write the better!
One thing has occurred to me (I've only just started to think about this so more will appear when I actually start to hack code):
make the file_list (and associated date, GPS, exif etc data) into an sqlite database see pi3d/pi3d_demos#41 (comment)
Beta Was this translation helpful? Give feedback.
All reactions