Can't load model

[nukadelic]

I have uploaded the coco model shard# files to remote server but can't seem to load it, please have a look at : https://codesandbox.io/s/y083p9nz39

[justadudewhohacks]

The final url your loading the model from seems to be wrong: https://uploads.codesandbox.io/uploads/user/38363f64-0d83-4bc1-946f-9f741badbeed/1YLA-coco_model-weights_manifest_uploaded/-weights_manifest.json.

I think you forgot the .json file ending of the manifest file, that you are trying to load:

let modelWeightsManifest = "https://uploads.codesandbox.io/uploads/user/38363f64-0d83-4bc1-946f-9f741badbeed/1YLA-coco_model-weights_manifest_uploaded.json";

[nukadelic]

The extra string in the end ( /-weights_manifest.json ) will be added to any file. just to make sure I attempted to referenced a picture URL, and it included this string as well. After some digging through the developer console I found the file where it's done: https://github.com/justadudewhohacks/tfjs-image-recognition-base/blob/master/src/common/getModelUris.ts#L2

without over complicating things, is it possible to load model from an imported JSON string? , like so:

import CocoWeights from '../model/coco_model-weights_manifest.json';
import CocoConfig from '../model/coco_model_config.json';
...
const net = new yolo.TinyYolov2(CocoConfig);
await net.load( CocoWeights );

This setup will produce the following error:
TinyYolov2.load - expected model uri, or weights as Float32Array

I got a similar setup working this way via tf.loadModel( tf.io.browserFiles( [ ... here is an example https://codesandbox.io/s/kor4m66poo?module=%2Fsrc%2FModelLoader.ts

[justadudewhohacks]

Maybe I did not fully understand, but as I said, you forgot the .json ending in your url. Thats why /-weights_manifest.json is appended to at the end, since what you give as input to the load function is not a valid uri to a json file.

To your second question, it will be possible to load the model from the json file as you suggested, once I publish the stuff I am currently working on.

[nukadelic]

Nope sorry, you had it right the first time, I didn't pay full attention to your answer and got carried away. Will be waiting for the update with anticipation.

[justadudewhohacks]

It's now possible to load your model from the weight map using net.loadFromWeightMap(weightMap) where weightMap is what you retreive from tf.io.loadWeights.