guichristmann / edge-tpu-tiny-yolo

Run Tiny YOLO-v3 on Google's Edge TPU USB Accelerator.
MIT License
102 stars 31 forks source link

Description of outputs? #9

Open snowzach opened 4 years ago

snowzach commented 4 years ago

I've written a tool called DOODS https://github.com/snowzach/doods that allows for offloaded object detection. It's written in Go and I would love to be able to work with this model (with the edge tpu)

Is there any documentation on the output format of this model? I am having a hard time following the code (I'm not a python guy) I see there are 2 outputs, 1, 13, 13, 255 and 1, 26, 26, 255

{"package": "detector.tflite", "name": "edgetpu2", "n": 0, "name": "Identity", "type": "UInt8", "num_dims": 4, "byte_size": 43095, "quant": {"Scale":0.08390676975250244,"ZeroPoint":230}, "shape": [1, 13, 13, 255]}
{"package": "detector.tflite", "name": "edgetpu2", "n": 0, "dim": 0, "dim_size": 1}
{"package": "detector.tflite", "name": "edgetpu2", "n": 0, "dim": 1, "dim_size": 13}
{"package": "detector.tflite", "name": "edgetpu2", "n": 0, "dim": 2, "dim_size": 13}
{"package": "detector.tflite", "name": "edgetpu2", "n": 0, "dim": 3, "dim_size": 255}
{"package": "detector.tflite", "name": "edgetpu2", "n": 1, "name": "Identity_1", "type": "UInt8", "num_dims": 4, "byte_size": 172380, "quant": {"Scale":0.08317398279905319,"ZeroPoint":214}, "shape": [1, 26, 26, 255]}
{"package": "detector.tflite", "name": "edgetpu2", "n": 1, "dim": 0, "dim_size": 1}
{"package": "detector.tflite", "name": "edgetpu2", "n": 1, "dim": 1, "dim_size": 26}
{"package": "detector.tflite", "name": "edgetpu2", "n": 1, "dim": 2, "dim_size": 26}
{"package": "detector.tflite", "name": "edgetpu2", "n": 1, "dim": 3, "dim_size": 255}
{"package": "detector", "name": "edgetpu2", "type": "tflite", "model": "models/quant_coco-tiny-v3-relu.tflite", "labels": 80, "width": 416, "height": 416}

Any help at all would be greatly appreciated!

songshan0321 commented 3 years ago

Same question here