FRCNN test.py won't create any bounding boxes on img

I’m trying to make a machine learning model on Car Damage Detection dataset, with annotations and 600 images.

I created the .CSV file containing the annotations in the format i.e. (filename, class, x1, y1, x2, y2), split the dataset into train.csv and test.csv

Then, to train the data using Faster-RCNN with ResNet-50 architecture, I cloned the GitHub repository containing Keras implementation of Faster-RCNN using this link.

I did everything what was told in the above link.

I was to run the “train_frcnn.py” in the repo to train the data. I trained the data at 50 epochs (Each epoch had 125 iterations, default epoch was 2000). Total ETA was 4 hours for 50 epochs.


(again ran the train file after knowing the epochs were default at 2000)

It created the “config.pickle” (contained some important data gathered from training) and a HDF5 file for saving the weights.

After it, I ran the “test_frcnn.py”, expecting the images with bounding boxes on the images.

Surprisingly, I get no images with bounding boxes on them (they were as it is as I stored them in the “test_images” folder)

I thought that the images after testing would be stored in “result_img” folder as mentioned in the “test_frcnn.py”, but found no such folder there.

I need help, regarding this.

© Copyright 2013-2019 Analytics Vidhya