SSD Mobilenet is not detecting any object.

[senthilnathansss]

Compiled and ran app in the Pixel-2 but no objects are getting detected. I know the coco trained SSD mobilenet works for Person and some objects like cup, monitor but this app detects nothing.

[fdchiu]

It should detect some objects. Are you using your own model or the model in the repo? Another thing to check is the camera, make sure the image is fed to ssd mobile net is correct image (size/exposure/etc).

David

On Tue, Oct 1, 2019 at 2:20 PM senthilnathansss notifications@github.com wrote:

Compiled and ran app in the Pixel-2 but no objects are getting detected. I know the coco training SSD mobilenet works for Person and some objects like cup, monitor but this app detects nothing.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/fdchiu/SNPECam/issues/3?email_source=notifications&email_token=ABKEYUJHRA2OFFZLAM2KJALQMO5KFA5CNFSM4I4O7MZKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HO67KHA, or mute the thread https://github.com/notifications/unsubscribe-auth/ABKEYUIYHTSZOGSTGSR3R5DQMO5KFANCNFSM4I4O7MZA .

[unitedsoft]

I tracked it looks like the fix is to change the below IF condition in run method of SNAPNet.java class.

//if (output.getKey().equals(mOutputLayer)) { if (key.startsWith("Postprocessor/BatchMultiClassNonMaxSuppression")) {

• SP

[fdchiu]

Thanks for the update! Everyone's network is different from others. IN your case your out put is "Postprocessor/BatchMultiClassNonMaxSuppression" so this change worked for you.

Just wondering if you set mOutputLayer to the above value, should the original statement still work?

David

On Thu, Nov 7, 2019 at 8:03 PM Saravanaperumal notifications@github.com wrote:

I tracked it looks like the fix is to change the below IF condition in run method of SNAPNet.java class.

//if (output.getKey().equals(mOutputLayer)) { if (key.startsWith("Postprocessor/BatchMultiClassNonMaxSuppression")) {

• SP

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/fdchiu/SNPECam/issues/3?email_source=notifications&email_token=ABKEYUIXNF6YSZWWN5ZOXIDQSTQHPA5CNFSM4I4O7MZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDOVDFY#issuecomment-551375255, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABKEYUNZ36WCUGZGFRXUYS3QSTQHPANCNFSM4I4O7MZA .

[unitedsoft]

Yes, you are right. Setting mOutputLayer should work. But haven't tested it yet. But thank again your work helped me and saved lot of time for me.

-- SP

[jeranio]

Hello

I was not getting any detections at all until I reversed the commented line as so:

//if (output.getKey().equals(mOutputLayer)) {
if (key.startsWith("Postprocessor/BatchMultiClassNonMaxSuppression")) {

the orginal file was

if (output.getKey().equals(mOutputLayer)) {
//if (key.startsWith("Postprocessor/BatchMultiClassNonMaxSuppression")) {

Then it starts to detect but it detects all kinds of things not even in the room like elephants. Is there a guide on training this model to be more accuate.

Thanks though for the good work

Running on SDM660 platform