cloud-annotations / object-detection-live-stream

🔴 Real-Time Object Detection on a Livestream with TensorFlow.js
MIT License
15 stars 2 forks source link

copy and pasted the model_web. I got this error. #2

Open noma6386-duplicate opened 4 years ago

noma6386-duplicate commented 4 years ago

models@0.1.1:1 Uncaught (in promise) Error: Tensor must have a shape comprised of positive integers but got shape [1917,]. at z (models@0.1.1:1) at models@0.1.1:1 at Array.forEach () at it (models@0.1.1:1) at Tn (models@0.1.1:1) at _n (models@0.1.1:1) at models@0.1.1:1 at models@0.1.1:1 at Kt.scopedRun (models@0.1.1:1) at Kt.tidy (models@0.1.1:1)


I tried other samples on your git.

I tried https://github.com/tensorflow/tfjs-models/tree/master/coco-ssd with the model. It was not succeed also. Help me out please.

bourdakos1 commented 4 years ago

I think you don't have enough training data or not enough labels, try to have at least 2 labels and 20 images per label. Otherwise this error pops-up somewhat randomly

noma6386-duplicate commented 4 years ago

I have 6 labels and used 200 images. I got my model trained with IBM cloud. cacli ver 1.3.2 (latest one) I got label, model.json and 6 bin files.

bourdakos1 commented 4 years ago

hmm yea that's definitely enough training data... Do you mind emailing me a screenshot of your training data and maybe share your bucket with me? bourdakos1@gmail.com

noma6386-duplicate commented 4 years ago

Sorry for the late response. I am living in other time zone. I email you with bucket and files.