|
1 | 1 | # Tensorflow_API-Custom_object_detection |
2 | 2 | A sample project to detect custom object using Tensorflow object detection API |
3 | 3 |
|
| 4 | + |
4 | 5 | ## Folder Structure |
5 | 6 | - Tensorflow_API-Custom_object_detection |
| 7 | + - pre_trained_model |
| 8 | + - *downloaded files for choosen pretrained model* |
6 | 9 | - dataset |
7 | 10 | - Annotations |
8 | | - - *Annotations for your images comes here* |
| 11 | + - *Annotations for your training images comes here* |
9 | 12 | - JPEGImages |
10 | 13 | - *all of your images for training comes here* |
11 | 14 | - testImages |
12 | 15 | - *all your images for testing comes here* |
13 | 16 | - lable.pbtxt |
14 | 17 | - train.record |
15 | | - - output |
| 18 | + - IG |
16 | 19 | - *inference graph will be saved here* |
17 | | - - train |
| 20 | + - CP |
18 | 21 | - *checkpoints will be saved here* |
19 | 22 | - eval.ipynb |
20 | 23 | - train.ipynb |
| 24 | + - *config file for the choosen model* |
| 25 | + |
21 | 26 |
|
22 | 27 | ## Steps |
23 | 28 |
|
24 | 29 | ###### Create folders |
25 | 30 | Create the folders following the structure given above |
26 | 31 |
|
27 | | -###### prepare train and test images |
28 | | -This repository contains train and test images for detection of "UE Roll" blue bluetooth speaker but I will highly recommend you to create your own dataset. Pick up an object you want to detect and take some pics of it with varying backgrounds, angles and distances. Sample of images used in this sample project is given below: |
| 32 | +###### Prepare train and test images |
| 33 | +This repository contains train and test images for detection of "UE Roll" blue bluetooth speaker but I will highly recommend you to create your own dataset. Pick up an object you want to detect and take some pics of it with varying backgrounds, angles and distances. Some of the sample images used in this sample project is given below: |
29 | 34 |
|
30 | 35 |  |
31 | 36 |
|
32 | | -Once you have captured pictures, transfer it to your PC and resize it to smaller size (given images have size of 605 x 454) so that your training will go smoothly without running out of memory. Now divide your captured images in to two chunks, one chunk for training and other for testing. Finally move training images in to *JPEGImages* folder and testing images in to *testImages* folder. |
| 37 | +Once you have captured pictures, transfer it to your PC and resize it to smaller size (given images have size of 605 x 454) so that your training will go smoothly without running out of memory. Now reaname (for better referencing later) and divide your captured images in to two chunks, one chunk for training(80%) and other for testing(20%). Finally move training images in to *JPEGImages* folder and testing images in to *testImages* folder. |
33 | 38 |
|
34 | 39 | ###### Label the data |
35 | | -Now its time to label your training data. We will be doing it using labelImg library. To download this library use [THIS LINK](https://github.com/tzutalin/labelImg). |
36 | | -Once you have labelImg library downloaded on your PC, run lableImg.py. Select *JPEGImages* directory by clicking on *Open Dir* and change your save directory to *Annotations* by clicking on *Change Save Dir*. Now all you need to do is to draw rectangles around the you are planning to detect after clicking on *Create RectBox* and save it so that Annotations will get saved as .xml file in *Annotations* folder. |
| 40 | +Now its time to label your training data. We will be doing it using [labelImg library](https://pypi.org/project/labelImg/). To download this library along with its dependencies use [THIS LINK](https://github.com/tzutalin/labelImg). |
| 41 | +Once you have labelImg library downloaded on your PC, run lableImg.py. Select *JPEGImages* directory by clicking on *Open Dir* and change your save directory to *Annotations* by clicking on *Change Save Dir*. Now all you need to do is to draw rectangles around the object you are planning to detect. You will need to click on *Create RectBox* and then you will get cursor to label the objects. After labeling image give the name for label and save it so that Annotations will get saved as .xml file in *Annotations* folder. |
37 | 42 |
|
38 | 43 |  |
39 | 44 |
|
40 | | -___________Under progress _____________ |
| 45 | +###### Setup Tensorflow models repository |
| 46 | +Now it time when we will start using Tensorflow object detection API so go ahead and clone it using following command |
| 47 | +``` |
| 48 | +git clone https://github.com/tensorflow/models.git |
| 49 | +``` |
| 50 | +Once you have cloned this repository, change your present working directory to models/reserarch/ and it to your python path. It you want to add it permanently then you will have to make change in your .bashrc file or you could do same temproarily for current session using following command: |
| 51 | +``` |
| 52 | +export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim |
| 53 | +``` |
| 54 | +You also need to run following command in order to get rid of string_int_label_map_pb2 issue (more details [HERE](https://github.com/tensorflow/models/issues/1595)) |
| 55 | +``` |
| 56 | +protoc object_detection/protos/*.proto --python_out=. |
| 57 | +``` |
| 58 | + |
| 59 | +###### Convert you data to Tensorflow record format |
| 60 | +In order to use Tensorflow API you need to feed data in Tensorflow record format. Thankfully tensorflow gives python script to convert Pascal VOC format dataset to tensorflow record format. Path for the file I mentioned in last line is given below: |
| 61 | +*models/research/object_detection/dataset_tools/create_pascal_tf_record.py* |
| 62 | +Now you have two options, either follow pascal VOC dataset format or modify the tesorflow script as per your need. I modified the script and I have placed same in this repository inside the folder named as *extra*. All you need to do is to take this script and replace original script with this. If you do so, you dont need to follow any specific format. |
| 63 | +After doing all this circus, there is one last thing is still remaining before we get our Tensorflow record file. You need to create a file for label map, in this repo its label.pbtxt, with dictionaly of label and id of items. Check label.pbtxt given in this repository to undestand the format, its pretty simple (Note: name of label should be same as what you had given while labeling object using labelImg). Now it time to create record file. From models/research as present working directory run following command to create Tensorflow record: |
| 64 | +``` |
| 65 | +python object_detection/dataset_tools/create_pascal_tf_record.py --data_dir=<path_to_your_data_directory> --annotations_dir=<name of annotations directory> --output_path=<path_where_you_want_record_file_to_be_saved> --label_map_path=<path_of_label_map_file> |
| 66 | +``` |
| 67 | +For more help run following: |
| 68 | +``` |
| 69 | +python object_detection/dataset_tools/create_pascal_tf_record.py -h |
| 70 | +``` |
| 71 | +An example path will be: |
| 72 | +``` |
| 73 | +Python object_detection/dataset_tools/create_pascal_tf_record.py --data_dir=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/dataset --annotations_dir==Annotations --output_path=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/dataset/train.record --label_map_path=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/dataset/label.pbtxt |
| 74 | +``` |
| 75 | + |
| 76 | +###### Training |
| 77 | +Now that we have data in right format to feed, we could go ahead with training our model. First thing you to do is to select the pretrained you would like to use. You could check and download pretrained model from [Tensorflow detection model zoo Github page](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md). Once downloaded, extract all file to the folder you had created for saving pretrained model files. Next you need to copy to *models/research/sample/configs/<your_model_name.config>* and paste in your project repo. You need to configure 5 paths in this file. Just open this file and search for PATH_TO_BE_CONFIGURED and replace it with path of required file. I used pretrained faster RCNN trained on COCO dataset and I have added modified config file (along with PATH_TO_BE_CONFIGURED as comment above lines which has been modified) for same in this repo. Now you are all set to train your model, just run following comand with models/research as present working directory |
| 78 | +``` |
| 79 | +python object_detection/legacy/train.py --train_dir=<path_to_the folder_for_saving_checkpoints> --pipeline_config_path=<path_to_config_file> |
| 80 | +``` |
| 81 | +An example will be |
| 82 | +``` |
| 83 | +python object_detection/legacy/train.py --train_dir=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/CP --pipeline_config_path=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/faster_rcnn_resnet101_coco.config |
| 84 | +``` |
| 85 | +Let it train till loss will be below 0.1 or even lesser. once you she loss is as low as you want then give keyboard interupt. Checkpoints will be saved in CP folder. Now its time to generate inference graph from saved checkpoints |
| 86 | +``` |
| 87 | +python object_detection/export_inference_graph.py --input_type=image_tensor --pipeline_config_path=<path_to_config_file> --trained_checkpoint_prefix=<path to saved checkpoint> --output_directory=<path_to_the_folder_for_saving_inference_graph> |
| 88 | +``` |
| 89 | +An example will be |
| 90 | +``` |
| 91 | +python object_detection/export_inference_graph.py --input_type=image_tensor --pipeline_config_path=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/faster_rcnn_resnet101_coco.config --trained_checkpoint_prefix=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/train/model.ckpt-1691 --output_directory=/Users/vijendra1125/Documents/tensorflow/object_detection/speaker_detection/IG |
| 92 | +``` |
| 93 | +**Bonus:If you want to train your model using Google colab then check out the train.ipynb file** |
| 94 | + |
| 95 | +###### Test the trained model |
| 96 | +Finaly its to time to check the result of all the hard work you did. All you need to do is to copy model/research/object_detection/object_detection_tutorial.ipynb and modfy it to work with you inference graph. A modified file is aready given as eval.ipynb with this repo, you just need to chnage path and the number of images you have given as test image. below is the result of the model trained for detecting "UE Roll" blue bluetooth speaker. |
| 97 | + |
| 98 | + |
| 99 | + |
41 | 100 |
|
42 | 101 |
|
0 commit comments