search

aws / amazon-kinesis-video-streams-parser-library

Amazon Kinesis Video Streams parser library is for developers to include in their applications that makes it easy to work with the output of video streams such as retrieving frame-level objects, metadata for fragments, and more.
Apache License 2.0
102 stars 52 forks source link

question about KinesisVideoRekognitionIntegrationExampleTest #28

Closed RileyJia closed 4 years ago

RileyJia commented 6 years ago

I am running this testcase, but the result video is the same as the input. It cannot draw box around the faces. I have create a collection and index Bezos's face. There is no other instructions about this test. I don't know where I did wrong.

bdhandap commented 6 years ago

@Rulovescoding Generally Rekognition takes some time to return the results for the detected faces causing the video to be delayed. If the detected faces are not returned within the playback duration of the video file then the video frames will be rendered without bounding boxes. Can you please increase the rekognition timeout to a higher value so that video frames will wait little bit longer until the rekognition results are returned?

Also a better experiment is to test with a live producer (camera) instead a short test file so that the video playback duration is much longer. Please take a look at KVS producer SDK demo applications to know how to send live stream to KVS.

RileyJia commented 6 years ago

I have tried to set the timeout to like 300 and 1000. It would return an error said about sleeping thread. The video frame viewer would be closed after the video display for 3 to 5 seconds. That’s definitely not the right result, I am not sure how big should I set the timeout variable.

On Wed, Aug 8, 2018 at 1:33 AM Babu Prasad notifications@github.com wrote:

@Rulovescoding https://github.com/Rulovescoding Generally Rekognition takes some time to return the results for the detected faces causing the video to be delayed. If the detected faces are not returned within the playback duration of the video file then the video frames will be rendered without bounding boxes. Can you please increase the rekognition timeout https://github.com/aws/amazon-kinesis-video-streams-parser-library/blob/master/src/test/java/com/amazonaws/kinesisvideo/parser/examples/KinesisVideoRekognitionIntegrationExampleTest.java#L64 to a higher value so that video frames will wait little bit longer until the rekognition results are returned?

Also a better experiment is to test with a live producer (camera) instead a short test file so that the video playback duration is much longer. Please take a look at KVS producer SDK demo applications https://github.com/awslabs/amazon-kinesis-video-streams-producer-sdk-cpp#sample-demo-applications-for-sending-video-to-kinesis-video-streams to know how to send live stream to KVS.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/aws/amazon-kinesis-video-streams-parser-library/issues/28#issuecomment-411138894, or mute the thread https://github.com/notifications/unsubscribe-auth/AZlr1k00OmjEweDcAPK6Unr_-QsTgaI2ks5uOc-EgaJpZM4VxsGl .

manabu-tsunokai commented 5 years ago

I have the same problem, no bounding box is drawn. I have tried to set the timeout 100 to 20000, no success. I have also tried a live producer using the camera of my macbook air, but the results were the same. The log shows that processRecords() of KinesisRecordProcessor is not called at all. The console of Kinesis Data Stream shows some data transfer is going on, but the chart of GetRecords.Bytes shows the values are always 0. I can successfully get the rekognition results including matched face through awscli using commands like "get-shard-iterator", "get-records". I would appreciate any help, thank you.

Region AP_NORTHEAST_1 MacOS 10.14.5 JDK 1.8.0_131

bdhandap commented 5 years ago

@manabu-tsunokai If GetRecords.Bytes is 0 then I think the RekognitionStreamProcessor is either not started or encountering some error. Can you do a describe on the stream process and check whether the status is RUNNING?

https://docs.aws.amazon.com/rekognition/latest/dg/streaming-video-troubleshooting.html

manabu-tsunokai commented 5 years ago

Thanks for the response. I confirmed that the stream processor is running. But after some trial and error, I solved the problem. I ran the test with a live producer, setting the argument of example.execute() to very high (such as 300L). Then it it started to show the bounding box 20 or 30 seconds after starting the test, although the frame rate is very very slow. Maybe the machine power is not sufficient to run the test smoothly. (I use macbook air early 2015)

MushMal commented 4 years ago

Resolving as stale. Please cut a new issue is this is still an issue