I'm creating a video sample by combining cropped-faced images using Deepface python library.
I saved the cropped faces in a list called cropped_faces. And it has saved all the frames successfully.
The list content of cropped_faces[]:
Finally, I am using cv2.write() method in a for loop to write the frames to the video. The code is as follows:
height,width,layers = cropped_faces[0].shape
print(height,width,layers)
fourcc = cv2.VideoWriter_fourcc(*'H264')
video = cv2.VideoWriter('/content/gdrive/MyDrive/Kaggle/Data/output.mp4', fourcc,1,(width,height))
for j in range(0,5):
video.write(cropped_faces[j])
video.release()
cv2.destroyAllWindows()
print("The video was successfully saved")
The output:
Although there are no errors in the code, the video saved in the location '/content/gdrive/MyDrive/Kaggle/Data/' is not playable. Only a size 258bytes video is created.
Can somebody please help me to find the reason and solve this issue, please?
Related
I am using Google cloud speech through Python and finding many transcriptions are inaccurate and missing several words. This is a simple script I'm using to return a transcript of an audio file, in this case 'out307.wav':
client = speech.SpeechClient()
with io.open('out307.wav', 'rb') as audio_file:
content = audio_file.read()
audio = speech.types.RecognitionAudio(content=content)
config = speech.types.RecognitionConfig(
enable_word_time_offsets=True,
language_code='en-US',
audio_channel_count=1)
response = client.recognize(config, audio)
for result in response.results:
alternative = result.alternatives[0]
print(u'Transcript: {}'.format(alternative.transcript))
This returns the following transcript:
to do this the tensions and suspicions except
This is very far off what the actual audio says (I've uploaded it at https://vocaroo.com/i/s1zdZ0SOH1Ki). The audio is a .wav and very clear with no background noise. This is worse than average, as in some cases it will get the transcription fully correct on a 10 second audio file, or it may miss just a couple of words. Is there anything I can do to improve results?
This is weird, I tried your audio file with your code and I get the same result, but, if I change the language_code to "en-UK" I am able to get the full response.
I'm working for Google Cloud and I created for you a public issue here, you can track there the updates.
I downloaded a jpg file from my ip cam.
but the file seems broken or something about its format.
I couldn't open it using safari or preview.app(macOS),
but it shows in chrome.
What I really trying to do is download it programmatically and show in UIImageView, but the image data is always nil.
I don't know much about image format and the question is strange,
so if you're willing to help me and see whats going on with that picture,
the image: broken jpg
and thanks for your time!
UPDATE 2017/03/30 :
Still haven't found the answer for how to decode motion jpeg frame.
From what I googled, the difference is DHT, but don't know how to add it to a frame.
As far as I know, there are few third-party libs like libjpeg-turbo, ffmpeg but haven't found an example.
If you have done this before and wrote in C or Objective-C, hope you can help me out!
Really want and need to know how!
Thanks!
Your file is an Motion JPEG file, not a JPEG image...
This does explain why browsers are able to open it and you can check this in vlc by looking at the codecs information :
Motion JPEG Video (MJPG)
or even through ffmpeg -i MeQ6p.jpg:
Stream #0:0: Video: mjpeg, yuvj422p(pc, bt470bg/unknown/unknown), 640x480, 25 tbr, 25 tbn, 25 tbc
So your problem lies in your files only, and this can probably be fixed by setting your ip-cam to save still JPEG images instead of MJPEG streams.
I just tried and can see the data gets fetched from your posted URL and got some 17750 bytes.
let imageURL = NSURL(string : "https://i.stack.imgur.com/MeQ6p.jpg")
let imageData = NSData(contentsOfURL: imageURLFromParse! as NSURL)
I'm trying to use the OSVR IR camera in OpenCV 3.1.
Initialization works OK.
Green LED is lit on camera.
When I call VideoCapture.read(mat) it returns false and mat is empty.
Other cameras work fine with the same code and VLC can grab the stream from the OSVR camera.
Some further testing reveals: grab() return true, whereas retrieve(mat) again returns false.
Getting width and height from the camera yields expected results but MODE and FORMAT gets me 0.
Is this a config issue? Can it be solved by a combination of VideoCapture.set calls?
Alternative Official answer received from the developers (after my own solution below):
The reason my camera didn't work out of the box with OpenCV might be that it has old firmware (pre-v7).
Work around (or just update firmware):
I found the answer here while browsing anything remotely linked to the issue:
Fastest way to get frames from webcam
You need to specify that it should use DirectShow.
VideoCapture capture( CV_CAP_DSHOW + id_of_camera );
For the past two weeks I am trying to find a proper way to read frames from .mts video file and process them in OpenCV. When .mts file is in 25p (25 fps progressive) format VideoCapture of OpenCV works fine for seeking video frames but when it is in 50i (25 fps interlaced) format VideoCapture of OpenCV can not properly decode it frame by frame.
(e.g. in a sample scenario when I get frame #1 and then read frame #300 and later read frame #1, it returns a corrupted image different from my previous read of frame #1) (i am using OpenCV 2.4.6)
I decided to replace video decoder part of the program.
I tried FFmpegSource2 but the problem of proper frame seeking for .mts was not resolved (most of the time FFMS_GetFrame function returns same output for several consecutive frames for 50i .mts file).
I also tried DirectShow. But IsFormatSupported method of IMediaSeeking for TIME_FORMAT_FRAME does not return S_OK for 50i .mts video file and it only supports TIME_FORMAT_MEDIA_TIME for this kind of video file. I have not tried myself but a friend said even using TIME_FORMAT_MEDIA_TIME for frame seeking will result in the same problem as above and I may not be able to jump back and forward to individual frames and read their data.
Now I am going to try gstreamer. I found sample method for linking gstreamer and openCV in the following link:
Adding opencv processing to gstreamer application
When I try to compile it in gstreamer 1.0, I get the following error:
error C3861: 'gst_app_sink_pull_buffer': identifier not found
I have included gst/gst.h, gst/app/gstappsink.h, gst/app/gstappsrc.h
Looked at the following help link and there was not gst_app_sink_pull_buffer function there too.
http://gstreamer.freedesktop.org/data/doc/gstreamer/head/gst-plugins-base-libs/html/gst-plugins-base-libs-appsink.html
I am using gstreamer 1.0 (v1.2.0) from gstreamer.freedesktop.org
May be gstreamer SDK from www.gstreamer.com (based on gstreamer 0.1) work for that, but I have not tried it yet and prefer to use gstreamer from gstreamer.freedesktop.org
I don't know where gst_app_sink_pull_buffer is defined. Anybody knows how I can compile the sample method provided for gstreamer 0.1 in
Adding opencv processing to gstreamer application for gstreamer 1.0?
Thank you in advance.
UPDATE 1: I am new to gstreamer. Now I know that have to port the sample method of Adding opencv processing to gstreamer application from gstreamer 0.1 to gstreamer 1.0. I replaced gst_app_sink_pull_buffer function with gst_app_sink_pull_sample and gst_sample_get_buffer. Have to work more on the other parts of the code and see if can open a desired frame from 50i .mts video file and process it with OpenCV.
UPDATE 2: I found a very good example at
http://gstreamer.freedesktop.org/data/doc/gstreamer/head/manual/html/section-data-spoof.html#section-spoof-appsink
And I easily replaced the part which saves the snapshot using GTk with functions that load frame data buffer to an OpenCV Mat. This program works fine for many video file types and I can grab frames of the video file in OpenCV Mat. But when the input video file is a 50i .mts video file, it returns the following errors and I can not read the frame data:
No accelerated IMDCT transform found
0:00:00.405110839 4632 0B775380 ERROR libav :0:: get_buffer() failed (-1 2 00000000)
0:00:00.405740899 4632 0B775380 ERROR libav :0:: decode_slice_header error
0:00:00.406401077 4632 0B7756A0 ERROR libav :0:: Missing reference picture
0:00:00.406705867 4632 0B7756A0 ERROR libav :0:: Missing reference picture
0:00:00.416044436 4632 0B7759C0 ERROR libav :0:: Cannot combine reference and non-reference fields in the same frame
0:00:00.416813339 4632 0B7759C0 ERROR libav :0:: decode_slice_header error
0:00:00.417725301 4632 0B775CE0 ERROR libav :0:: Missing reference picture
The step by step debug shows that "No accelerated IMDCT transform found" appears after running
ret = gst_element_get_state( pipeline, NULL, NULL, 5 * GST_SECOND );
and google search shows that I can ignore it as a warning.
All of the other errors emerge just after running
g_signal_emit_by_name( sink, "pull-preroll", &sample, NULL );
I have no idea how to resolve this issue? I have already played this .mts file in another example using playbin and gstreamer can play this .mts video file well when I use playbin.
My understanding currently is that:
CameraUI
I can use the CameraUI to access the built in camera for MediaType.VIDEO and that delegates to the built-in video camera app and lets me record a video. My app does that now.
When I stop recording and click the "Use" button, I am returned to my app and theoretically I have a valid MediaPromise.
iOS does -not- provide a valid/usable url/filename to the recorded video (or to photos) and so I would have to use a Loader to bring-in/use/access the 'recorded' video... AND... iOS does not actually create a file anywhere on the device, most importantly, in the Camera Roll where one would expect by the normal behavior when uses the system native camera/video app.
The documentation says that the Loader can load various image types and SWFs but nothing about video data, so I conclude from that that I cannot actually use the CameraUI to generate a valid MediaPromise that I can then pass to a Loader class or similar to read in the information created by the system camera and then manipulate (upload, save to applicationStorageDirectory, and/or display in one of the two video player components available in the API).
CameraRoll
I can have video entities in the iOS Camera Roll but the AS3/Air3.5 CameraRoll class won't let me view/access/reference them in any way.
Normal File I/O
All my attempts to use the Air3.5 File classes to browse to the storage location of the iOS Camera Roll have been rebuffed.
------- Questions -------
Am I correct in believing that there is a way to take video but no way to use the video that's been captured. (No way to use the resulting MediaPromise successfully).
I believe you can take video and access it using Android, but there's nothing in the documentation that says that you cannot using iOS.
Am I correct in believing that iOS sandboxes apps so that they cannot browse to video/photo storage using standard File I/O, but only through the apparently non-workable means I've tried (CameraUI & CameraRoll)
Am I wrong to think that these should be rather obvious NEEDS that one can achieve using the XCode Objective C++ etc route but the AIR Mobile Framework does not allow either because of Apple blocking functionality or because Adobe has failed to meet reasonable expectations?
One item of ironic note to convey. If I use the iOS system camera app to record a video, a thumnail of that video then appears in the Gallery/Camera Roll, and of course, I can share it or view it, or whatever... If I use AIR's CameraRoll.browseForImage(), provided I haven't used the camera to take another image, when it shows me the folder where the pictures are stored, the folder icon uses the thumbnail of the last object added... in this case, the video I took, but if I then enter the folder, the video cannot be found. It's teasing us. It knows it's there, but it is apparently forbidden fruit.
I can't answer all your questions, so this entry may not be acceptable, but I found this page while searching a solution for some the problems you described and thought that someone else may find this answer (partially) useful.
To save the movie you just took you need to open and read the data from the promise.
The iOS won't save the file anywere, so the MediaPromise.file is always null.
This is my solution to the problem:
private var camera:CameraUI;
private var dataInput:IDataInput;
public function recordVideo():void
{
// Start the camera and ask for a video
camera = new CameraUI();
camera.addEventListener(MediaEvent.COMPLETE, onCameraComplete);
camera.launch(MediaType.VIDEO);
}
private function onCameraComplete(event:MediaEvent):void
{
// event.data is a MediaPromise and MediaPromise.open() returns a IDataInput
// Let's cast it to a dispatcher and check when it's complete
dataInput = event.data.open();
var dispatcher:IEventDispatcher = IEventDispatcher(dataInput);
dispatcher.addEventListener(Event.COMPLETE, onDataInputComplete);
}
private function onDataInputComplete(event:Event):void
{
// We can do whatever we want with the data, so we'll store it in a File
var file:File = new File();
var bytes:ByteArray = new ByteArray();
var stream:FileStream = new FileStream();
// Reading the data from the opened MediaPromise
dataInput.readBytes(bytes);
stream.open(file, FileMode.WRITE);
stream.writeBytes(bytes, 0, bytes.bytesAvailable);
stream.close();
}
Also, I'm still looking for a way to put the movie in the CameraRoll