Closed snowzach closed 4 years ago
@r1cht3r are you able to give me a hint to what the output of that racoon model is? It just shows that it outputs 1 tensor that's uint8 and 5 bytes. I can't seem to figure out how to decode it.
Here's the tflite_metadata.json
that was provided with the tflite file from GCP, looks like just scores are outputted. Could the supportedTfVersions
be an issue?
{
"batchSize": 1,
"imageChannels": 3,
"imageHeight": 224,
"imageWidth": 224,
"inferenceType": "QUANTIZED_UINT8",
"inputTensor": "image",
"inputType": "QUANTIZED_UINT8",
"outputTensor": "scores",
"supportedTfVersions": [
"1.10",
"1.11",
"1.12",
"1.13"
]
}
Okay, I added support for this model type. This one doesn't actually provide boxes, it just provides a match percentage for every label.
@snowzach working great here, thank you!
Add support for object detection models that only return a score. This should return the confidence with the entire image selected [0,0,1,1] for the bounding match box.