tensorflow fined tune model Fast-RCNN does not show bounding box - python

I fined tuned both SSD Mobilenet and Fast-RCNN on the same dataset. Both models ran training and inference without any error. But Fast-RCNN fined tuned model does not show any bounding boxes. So I tried with one training image to see if the trained model is able to draw a bbox on an image that it was trained from. But it shows nothing. Where should I start looking for debugging ?

Related

YOLOV5 is not giving prediction on live webcam?

Actually i have trained a yolov5 model with traffic sign dataset diamentions (1360,800) with size of each image is 600KB but when i do real-time prediction on laptop camera its note able to predict on those signs. The diamention of webcame image are (600,450) with size 280KB does this problem due to the size or diamention of the image.
One thing to keep in mind i have no GPU,CPU in local pc i have trained model on colab. and its working fine on image for test in dataset with high size and diamentions.
This YOLOV5 trained model is working on test data of their own dataset but its not working on my captured image or live webcame of my pc.

How can I deploy a trained CNN model to production on a ARM?

I have trained a CNN model with Keras for semantic segmentation of craneal images and saved the weights and this trained model.
Now, I want to put it into production on a microprocessor. The pipeline of the process in the micro involves reading an image from a sensor and using it as input for the CNN model (U-Net). Then, the resulted binary image is used as a mask for an area of interest from which a variable is measured. Finally, a number is given as a result.
So, is it possible to load a trained model on a microprocessor? And if so, how?
Which features should have the microprocessor in order to work with CNN models?
Thanks in advance!

If Keras CNN model is trained on 28x28 images, how can I accurately predict images of higher resolutions?

I am trying to train an OCR model using CNN's and have an EMNIST dataset with letters for training the model to recognize handwritten English letters.
My question is, if I train the CNN with input image shape of 28x28, then during prediction, do I have to resize the image to be predicted to 28x28, or something else?
I am asking this because I originally trained a CNN model on 90x90 images and I am achieving a decent 93.4% accuracy on them. But when I resize similar images to 90x90 for prediction, the model predicts way out of the correct label.
I don't know what I am missing here, so all your guidance is duly appreciated.

Calculating the mAP from scores obtained in Tensorflow object detection API

I am actually working on a server where I run the Tensorflow OD API to train the model for my custom dataset. So, I divide my images into training, validation and test sets and run the train.py on training and validation sets. Next I run the inference using exported model checkpoint and frozen graph on my test images.
Now, my question is when I run the inference as provided in the example , I get an output dict with detection scores, number of detections, detection classes, detection masks etc for each image, so from these outputs how do I calculate the mAP for my test set?
Any guidance in this direction will be really helpful, thanks in advance.
You can use COCO's API for calculating COCO's metrics withing TF OD API. See this.
TF feeds COCO's API with your detections and GT, and COCO API will compute COCO's metrics and return it the TF (thus you can display their progress for example in TensorBoard). mAP#0.5 is probably the metric which is most relevant (at it is the standard metric used for PASCAL VOC, Open Images, etc), while mAP#0.5:0.95 is a much more difficult one localization-wise.

face_recognition knn algorithm: Is it possible add more faces in the generated .clf file without have to retrain everything again?

I'm using the KNN method from face_recognition library to train some faces, this method generate a .clf file with a trained model used for recognition. Can I add more faces in this file without have to train everything again?
Parallel thread in GitHub
It is convenient to use the transfer learning method, building your classification on top of a pre-trained model.
In this case, you'll use pre-train weights from a consistent model, and train your dataset on top of that, changing the network only at classification layers.
For facial (and emotion) recognition, VGG-16 models are indicated, and you'll have better accuracy.
For more information, please refer to this link:
https://github.com/anujshah1003/Transfer-Learning-Face-Recognition-using-VGG-16

Categories

Resources