google-coral / tflite

Examples using TensorFlow Lite API to run inference on Coral devices
https://coral.withgoogle.com
Apache License 2.0
182 stars 68 forks source link

Coral Detection Demo wrong tflite file #24

Closed mwkldeveloper closed 3 years ago

mwkldeveloper commented 4 years ago

In the detection example, install_requirements.sh indicate wrong tflite file, which is 404 not found:

(cd "${MODEL_DIR}"
curl -OL "${TEST_DATA_URL}/mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite" \
     -OL "${TEST_DATA_URL}/mobilenet_ssd_v2_coco_quant_postprocess.tflite" \
     -OL "${TEST_DATA_URL}/coco_labels.txt")

It cause error when try the example: image

Correct tflite file should be: image https://github.com/google-coral/edgetpu/tree/master/test_data

ssd_mobilenet_v2_coco_quant_postprocess.tflite ssd_mobilenet_v2_coco_quant_postprocess_edgetpu.tflite

Namburger commented 4 years ago

Thanks, will get this fixed!

Namburger commented 4 years ago

@yaotechiu sorry, this is already fixed for us internally, will push it out in next release

dudqls1994 commented 4 years ago

@yaotechiu sorry, this is already fixed for us internally, will push it out in next release image

I use /mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite this model! Is this problem is solved??? I faced the same error.

Namburger commented 4 years ago

@dudqls1994 the solution is to change mobilenet_ssd to sss_mobilenet in the installation script and when running the model. We are in a middle of a big release and all these changes are expected in mid q4

geogeorgiev commented 4 years ago

@yaotechiu sorry, this is already fixed for us internally, will push it out in next release image

I use /mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite this model! Is this problem is solved??? I faced the same error.

I solved this issue. It happens when downloading the .tflite from a device where you are not logged in to github (e.g. raspberry pi). In this case the file you download is an html text file (hence <!D, part of <!Doctype>...) instead of the buffer file holding the model.

Solution: I just downloaded the .tflite model file on my workstation where I am logged in and transferred the file to the device where the demo is run.

manoj7410 commented 3 years ago

This has been fixed. Feel free to reopen this thread if you still face the issue.