lars76 / object-localization Goto Github PK

Hi Lars,

I have tried running the code on a dataset for 100 epochs. After around 50 epochs, the error reduced from 2000(initially) to 25 but it stopped suddenly. How many times I have tried, I end up with same result. I couldn't figure out why it is stopping in between.

Could you help me in resolving this issue?

The exact error is placed in a screenshot. It is saying the val_iou didn't improve from 0.

Also, how to use GPU to run the python script. Currently it has multi-threading options which use CPU but I have a GPU supported system. How to change the access?

Thanks in advance lars
Srinath

I downloaded the data-set but it kips saying No such file or directory: 'train.csv'

Hello Lars, I would consider myself as a beginner in CNNs and would be grateful if you could assist with the problem I have. I have downloaded dataset and put .jpg and .xml files in the dataset folder. I have decided to put 4 instances for now to test the localization but it didn't work. train.csv and
validation.csv files that are generated at the beginning when running generate_dataset.py are empty. Do you have any suggestions on why it won't work for me?

Kind regards, Arseniy

Since the xml files have been generated, then can directly pass in to the neural network and train it? Or else need to manipulate them into array or csv file beforehand?

I see these two image files are corrupted each time I download them from the webpage:
beagle_116.jpg
chihuahua_121.jpg

When I run generate_dataset.py, I get this error. How can I fix this?

File "generate_dataset.py", line 115, in
main()
File "generate_dataset.py", line 62, in main
print("class {}: {} images".format(output[j-1][-2], i))
TypeError: unsupported operand type(s) for -: 'tuple' and 'int'

runfile('/Users/huinaxu/Desktop/object-localization-master/evaluate_performance.py', wdir='/Users/huinaxu/Desktop/object-localization-master')
Reloaded modules: generate_dataset, train_model
IoU on training data
2949/29499497/294911/294916/294920/294925/294930/294934/294939/294944/294949/294954/294958/294963/294968/294971/294976/294981/294986/294991/294996/2949101/2949106/2949110/2949114/2949120/2949123/2949126/2949132/2949135/2949138/2949143/2949146/2949149/2949153/2949158/2949163/2949168/2949171/2949174/2949177/2949180/2949183/2949186/2949189/2949193/2949198/2949203/2949208/2949213/2949217/2949222/2949227/2949230/2949233/2949236/2949239/2949242/2949247/2949252/2949257/2949262/2949266/2949271/2949276/2949281/2949284/2949287/2949290/2949293/2949296/2949299/2949302/2949306/2949311/2949315/2949319/2949323/2949328/2949333/2949338/2949343/2949346/2949349/2949354/2949358/2949362/2949366/2949370/2949373/2949376/2949380/2949383/2949386/2949390/2949395/2949400/2949404/2949410/2949415/2949419/2949424/2949427/2949431/2949435/2949439/2949443/2949447/2949451/2949456/2949461/2949466/2949471/2949476/2949481/2949486/2949491/2949495/2949499/2949504/2949509/2949513/2949517/2949521/2949524/2949527/2949530/2949534/2949539/2949543/2949547/2949553/2949558/2949563/2949568/2949572/2949577/2949582/2949587/2949592/2949596/2949601/2949608/2949612/2949617/2949622/2949627/2949632/2949637/2949642/2949646/2949649/2949653/2949659/2949664/2949669/2949674/2949679/2949682/2949685/2949688/2949691/2949695/2949699/2949703/2949708/2949712/2949715/2949719/2949723/2949727/2949732/2949737/2949741/2949745/2949749/2949753/2949758/2949763/2949767/2949771/2949775/2949778/2949782/2949785/2949788/2949791/2949796/2949801/2949806/2949810/2949814/2949818/2949822/2949827/2949832/2949837/2949841/2949847/2949851/2949855/2949860/2949864/2949868/2949873/2949878/2949882/2949886/2949890/2949894/2949900/2949904/2949908/2949912/2949915/2949920/2949924/2949928/2949932/2949937/2949942/2949947/2949951/2949955/2949959/2949963/2949967/2949970/2949972/2949976/2949980/2949984/2949988/2949992/2949996/29491000/29491004/29491008/29491013/29491017/29491021/29491025/29491029/29491033/29491037/29491041/29491046/29491050/29491054/29491058/29491062/29491066/29491070/29491073/29491076/29491079/29491084/29491088/29491091/29491095/29491099/29491102/29491106/29491110/29491114/29491118/29491122/29491126/29491130/29491134/29491138/29491142/29491146/29491150/29491154/29491158/29491162/29491166/29491170/29491174/29491178/29491182/29491186/29491190/29491194/29491198/29491202/29491205/29491210/29491214/29491219/29491223/29491227/29491231/29491235/29491240/29491244/29491248/29491252/29491256/29491260/29491264/29491268/29491272/29491276/29491280/29491284/29491294/29491311/29491336/29491367/29491411/29491478/29491582/29491739/29491996/29492402/2949
Avg IoU: 0.5732865147951192
Highest IoU: 0.9722222222222222
Lowest IoU: 0.0

IoU on validation data
737/737103/737
Avg IoU: 0.5246796910795323
Highest IoU: 0.9227614490772386
Lowest IoU: 0.0

Trying out unscaled image
images/Egyptian_Mau_167.jpg
Traceback (most recent call last):

File "", line 1, in
runfile('/Users/---/Desktop/object-localization-master/evaluate_performance.py', wdir='/Users/---/Desktop/object-localization-master')

File "/anaconda3/lib/python3.6/site-packages/spyder/utils/site/sitecustomize.py", line 705, in runfile
execfile(filename, namespace)

File "/anaconda3/lib/python3.6/site-packages/spyder/utils/site/sitecustomize.py", line 102, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)

File "/Users/---/Desktop/object-localization-master/evaluate_performance.py", line 119, in
main()

File "/Users/h---/Desktop/object-localization-master/evaluate_performance.py", line 108, in main
pred = predict_image(path, model)

File "/Users/---/Desktop/object-localization-master/evaluate_performance.py", line 42, in predict_image
if im.shape[0] != IMAGE_SIZE:

AttributeError: 'NoneType' object has no attribute 'shape'

What model did you use, is that Faster_RCNN or other ?I was trying to do object localization using Faster_RCNN and i only prepare datasets with annotation but i don't know how to extract features from annotation file as an input to fed to region proposal layer?Can you help please?

For example 1, if no ground truth available in the image, should I use all 0 to represent the gt bbox?

Hi, I'm using your code as a jumping off point for learning about localiser development, and was just a bit confused as to what STD and MEAN are explicitly used for? I think it looks like some kind of normalisation operation, but not 100% sure. Any info you can give would be greatly appreciated.
Cheers.

HHH-MBP:object-localization-master1 huinaxu$ python3 example_1/train_model.py
/Users/hhh/anaconda3/lib/python3.6/site-packages/h5py/init.py:36: FutureWarning: Conversion of the second argument of issubdtype from float to np.floating is deprecated. In future, it will be treated as np.float64 == np.dtype(float).type.
from ._conv import register_converters as _register_converters
Using TensorFlow backend.
Traceback (most recent call last):
File "example_1/train_model.py", line 130, in
main()
File "example_1/train_model.py", line 125, in main
model = create_model(IMAGE_SIZE, ALPHA)
File "example_1/train_model.py", line 75, in create_model
model = MobileNetV2(input_shape=(size, size, 3), include_top=False, alpha=alpha)
File "/Users/hhh/anaconda3/lib/python3.6/site-packages/keras_applications/mobilenet_v2.py", line 355, in MobileNetV2
expansion=1, block_id=0)
File "/Users/hhh/anaconda3/lib/python3.6/site-packages/keras_applications/mobilenet_v2.py", line 461, in _inverted_res_block
in_channels = inputs._keras_shape[-1]
AttributeError: 'Tensor' object has no attribute '_keras_shape'

Reloaded modules: train_model
Traceback (most recent call last):

File "", line 1, in
runfile('C:/Users/huixu/Desktop/object-localization-0812/train_model.py', wdir='C:/Users/hhh/Desktop/object-localization-0812')

File "C:\Localdata\Python3\lib\site-packages\spyder\utils\site\sitecustomize.py", line 705, in runfile
execfile(filename, namespace)

File "C:\Localdata\Python3\lib\site-packages\spyder\utils\site\sitecustomize.py", line 102, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)

File "C:/Users/hhh/Desktop/object-localization-0812/train_model.py", line 122, in
main()

File "C:/Users/hhh/Desktop/object-localization-0812/train_model.py", line 118, in main
train(model, EPOCHS, BATCH_SIZE, PATIENCE, TRAIN_CSV, VALIDATION_CSV, MEAN, STD)

File "C:/Users/hhh/Desktop/object-localization-0812/train_model.py", line 102, in train
train_datagen = DataSequence(train_csv, batch_size, mean, std)

File "C:/Users/hhh/Desktop/object-localization-0812/train_model.py", line 47, in init
for index, (path, x0, y0, x1, y1) in enumerate(reader):

ValueError: not enough values to unpack (expected 5, got 0)

Is it possible to infer trained model in openvino or even in opencv dnn module, as the openvino supports standard object detection models like, mobilenet with SSD, yolo, mask-rccn,etc. and simple classification models with respective neural network architecture(ex. inception,mobilenet,etc), so is it possible to convert h5 keras models to convert into native tensorflow frozen graph and infer it on openvino.

Thanks your code,it is very cool.But now I want get pic cat or dog localization and classname,what should I do?

def iou(y_true, y_pred):
    xA = K.maximum(y_true[...,0], y_pred[...,0])
    yA = K.maximum(y_true[...,1], y_pred[...,1])
    xB = K.minimum(y_true[...,2], y_pred[...,2])
    yB = K.minimum(y_true[...,3], y_pred[...,3])

    interArea = (xB - xA) * (yB - yA)

    boxAArea = (y_true[...,2] - y_true[...,0]) * (y_true[...,3] - y_true[...,1])
    boxBArea = (y_pred[...,2] - y_pred[...,0]) * (y_pred[...,3] - y_pred[...,1])
    return K.clip(interArea / (boxAArea + boxBArea - interArea + K.epsilon()), 0, 1)

in the above function, y_true and y_pred are directly coming from the train data and the model respectively - which are scaled coordinates. Shouldn't the iou function convert them back to xmin, ymin, xmax,ymax and then calculate the iou?

Have you any solution for this problem?The loss start to 17 and not decrement.

Hi,

I've run the test.py script, but I got an error:

Unable to open file (unable to open file: name = 'model-0.29.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)

I looked into my directory, but there isn't a file named like that. But I do have the files: 'model-0.44.h5' till 'model-0.51.h5'. Even when I changed the 'model-0.29.h5' into 'model-0.44.h5' in WEIGHTS_FILE in test.py and run again, it isn't working. Why this isn't working and how can I fix this? Should I have the file 'model-0.29.h5' into my directory? Should I change: save_best_only=TRUE into FALSE in the train.py script?

Hi, I prepared the dataset and in starting the training I have this error. I am trying to understand what may come from but I'm having difficulties.

File "train.py", line 175, in getitem
batch_boxes[i, floor_y, floor_x, 0] = (y1 - y0) / image_height
IndexError: index 7 is out of bounds for axis 1 with size 1

Hi. thanks for your amazing work. this model works very nice. But i have a question. when i feed the model with an image with two or three cats, I have only one box and its accuracy is very low. How can I modify the model for multi detection? what parts of the train_model.py or evaluate_performance.py should be changed for that? Your response will be highly appreciated.