Comments (4)
The output layer structure of EfficientDet and MobileNetV2-SSD is different. Also, the model I committed to the coco folder is the version of the model I didn't deliberately add postprocess to.
VOC
COCO
Models that do not come with a postprocess in the final layer will have to write their own decoding process. For example, you may find my article below helpful.
Post-Porcess by Python - Qiita
The sample logic below performs faster post-processing.
https://github.com/PINTO0309/PINTO_model_zoo/blob/master/051_East_Text_Detection/text_detection_video_tflite.py
from pinto_model_zoo.
There seems to be a bug in the official tflite export logic and tf-nightly==2.4.0-dev20200923 from a month ago. It doesn't quantize correctly and doesn't seem to return the correct result. I may have to quantize it again using the latest automl repository program. I'll try again when I have time.
https://github.com/google/automl.git
from pinto_model_zoo.
When I clone the latest master and convert tflite it seems to return some value. But it doesn't seem to be a correct value.
- tf-nightly==2.4.0-dev20200923
python3 model_inspect.py \
--runmode=saved_model \
--model_name=efficientdet-d0 \
--ckpt_path=efficientdet-d0 \
--saved_model_dir=savedmodeldir \
--tflite_path=efficientdet-d0.tflite
from pinto_model_zoo.
@PINTO0309 Thank you for looking into this. I was trying to figure out what everything meant, but had no idea. I have also looked at: https://github.com/PINTO0309/PINTO_model_zoo/blob/master/006_mobilenetv2-ssdlite/01_coco/mobilenetv2ssdlite.py. I see that you are using the post processing version. I also tried that but it seems like len(output_details) == 2
. Its almost as if there was no postprocessing? This is what the output_details looks like:
[{'name': 'raw_outputs/box_encodings', 'index': 315, 'shape': array([ 1, 1917, 4]), 'shape_signature': array([ 1, 1917, 4]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'raw_outputs/class_predictions', 'index': 316, 'shape': array([ 1, 1917, 91]), 'shape_signature': array([ 1, 1917, 91]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]
from pinto_model_zoo.
Related Issues (20)
- bad results for 342_ALIKE HOT 1
- Aborted (core dumped) for full quantized tinyhitnet model
- 091_gaze-estimation-adas-0002 network HOT 1
- Release new 303_FAN with heatmaps HOT 3
- dataset HOT 1
- 410_FaceMeshV2 quantized tflite models are not functional HOT 1
- 053_BlazePose / 058_BlazePose_Full_Keypoints source HOT 3
- How to retrain simple MLP (palm_detection_full_inf_post_192x192.onnx) model with custom hand dataset ? HOT 2
- How to retrain simple MLP (palm_detection_full_inf_post_192x192.onnx) model with custom hand dataset ? HOT 2
- MiDaS conversion script to onnx
- Mismatch between BlendshapeV2 TensorFlow and TFLite Model
- 425_Gold-YOLO-Body-Head-Hand cannnot execute demo_goldyolo_onnx_image.py HOT 1
- How to increase batch size when using post_process_gen_tools HOT 5
- Add DepthAnything-V2
- Documentation on using 403_trt_pose_hand on non trt systems? HOT 2
- How To Convert Mediapipe BlazePose For Coral TPU HOT 8
- convert movenet tflite failed
- CREStereo "next" onnx models for tensorrt HOT 1
- Midas 3.1 Conversion Script HOT 2
- 032_FaceMesh Score Normalization
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from pinto_model_zoo.