How you gather/label the data is maybe the least covered - important step in machine learning. 95% of learning material assumes you won't use it in the real world and therefore never teach you how to use real-world data. That is why I like this video a lot - it includes this part(not in full but enough for me).
I am labeling images with LabelImg and with every picture i have, i get 1 xml file. How do i put all those XML files in just one CSV to load it into my model. Here you use csv file from gs://cloud, how can I make a local csv file with the images I have?
Custom object detection is working but when I click the camera button .... nothing happens. Any clues ... I am using a Pixel 4a and would like to use the camera
Can we have evaluation function for object detection models in tensorflow zoo 2 also. Like for every class , the eval file only show mAP for all , not for every object individually.
Followed this tutorial for a custom dataset where images were stored in a local path. Doesn't work. I'm getting an error while importing the CSV of the dataset that says, "Image format not JPEG", despite the images being in the said format. Have tried doing it for both, .jpg and .jpeg extensions but still, the error persists.
You need to check the folder where your images are stored. You will find that not all of them have the .jpg extension. Just deleted the inappropiate images from that folder and it should work
@@wintoncape7995 Thanks, had tried that too which didn't work. However, I have solved this issue successfully by writing a Python script that converts all the image data into JPG format programmatically using PIL.
This might seem like a weird question but at what point when using Model Maker do you preference the headless model you're using? is it when you select your Model Architecture (EfficentDet-Lite2 etc.)?
4:59 : spec = model_spec.get('efficientdet_lite2') for efficientdet_lite2 , if you want efficientdet_lite1 you just change the name of model spec = model_spec.get('efficientdet_lite1')
I labeled my dataset on roboflow, I've saved it on my drive. then I tried to load it from my drive when I get into step 3 to train the model I get an attribute error that goes like: 'AttributeError: 'NoneType' object has no attribute 'label_map'.. please can somebody help me?
the label map "name of class to classify/detect" is missing. maybe in the csv exported from roboflow you have to organise as explained in the minute (2:42) because roboflow export it with a different format.
I've changed the number of detections in the efficientnet model (tflite_max_detections) before I train and also in Android Studio with (.setMaxResults) in the ObjectBuilder method but still the maximum number of detections is stuck at 25. Anyone any solutions?
I am labeling images with LabelImg and with every picture i have, i get 1 xml file. How do i put all those XML files in just one CSV to load it into my model?? im kinda confused here :x And in the CSV file i have to put the Train, Test or validation? Thanks
Well, you need not create a CSV for it. It supports PascalVOC too. However, I'm facing one error. While loading the data, an error occurs that says, "Image format not JPEG", despite the images being of the said format. Try your luck and let me know if you're successful.
@@nitintiwari7004 I tried to put all files on Google Drive, and It is succeeded without problem. If uploading to Google drive is not bother you, it is good way.
@@ningbo2197 Thanks for the suggestion. However, I resolved the issue. I pre-processed all the images by programmatically converting them to RGB using PIL library before feeding to the model and it worked just perfect.
@@nitintiwari7004 can you please share the code ? I have tried to feed my data from the drive and it is generating some random xxxx.jpeg path and gives error that file does not exist. I have only 225 images..
Please someone teach to me, how to stop training on the way to avoid overfitting? I cancel object_detector.create() by pushing run button, then "model" variable is not recognized on model.evaluate(test_data).
Instead of making a video on model training & deploying app, you should have just concentrated on how to train a tflite model. This video is jack of all and master of none. Please make a noob-friendly tutorial for a noob like me who wants to learn TF
How you gather/label the data is maybe the least covered - important step in machine learning. 95% of learning material assumes you won't use it in the real world and therefore never teach you how to use real-world data. That is why I like this video a lot - it includes this part(not in full but enough for me).
that optional testing part is what i needed
I am labeling images with LabelImg and with every picture i have, i get 1 xml file. How do i put all those XML files in just one CSV to load it into my model.
Here you use csv file from gs://cloud, how can I make a local csv file with the images I have?
Good tutorial sir! I salute you! Do you have any guide on how you load dataset from local computer?
I'm getting this error "required broadcastable shapes [Op:Mul]". on "model.evaluate_tflite"
None of the links in the description working
Custom object detection is working but when I click the camera button .... nothing happens. Any clues ... I am using a Pixel 4a and would like to use the camera
Using colab, the first "required packages" takes so much time and space it actually fills up the Disk lol
whats up with that
same, it's outdated 💀
Can we have evaluation function for object detection models in tensorflow zoo 2 also. Like for every class , the eval file only show mAP for all , not for every object individually.
Links don't work anymore.....
Followed this tutorial for a custom dataset where images were stored in a local path. Doesn't work. I'm getting an error while importing the CSV of the dataset that says, "Image format not JPEG", despite the images being in the said format. Have tried doing it for both, .jpg and .jpeg extensions but still, the error persists.
Change the image to .png and try - maybe you're data set format is mixed jpeg and png .
You need to check the folder where your images are stored. You will find that not all of them have the .jpg extension. Just deleted the inappropiate images from that folder and it should work
@@wintoncape7995 Thanks, had tried that too which didn't work. However, I have solved this issue successfully by writing a Python script that converts all the image data into JPG format programmatically using PIL.
Do you have any tutorial in recognizing images with accuracy and voice output?
This might seem like a weird question but at what point when using Model Maker do you preference the headless model you're using? is it when you select your Model Architecture (EfficentDet-Lite2 etc.)?
4:59 : spec = model_spec.get('efficientdet_lite2') for efficientdet_lite2 , if you want efficientdet_lite1 you just change the name of model spec = model_spec.get('efficientdet_lite1')
@@lotfimerad658 HOW MANY OBJECTS CAN I DETECT AT LEAST?
Hello sir, i want to ask. How we can get the validation accuracy during training?
variable validationnya diisi bg sebelum training, pake y_test,y_train
Thanks a lot for the great video!
I am trying to show thew results in listview instead of on image with boundingbox for testing purpose. how do i do that?
Can someone explain
"What samples are used to decide when to stop training to avoid overfitting."
Have you found the answer?
can u add a pick image from gallery module for this project?
I am getting an error like Disk is full , but I have 1TB free space in my SSD , how to solve this issue ?
Helll anyone can u plz answer this question what samples are used to decide when to stop training to avoid overfitting
I labeled my dataset on roboflow, I've saved it on my drive. then I tried to load it from my drive
when I get into step 3 to train the model I get an attribute error that goes like: 'AttributeError: 'NoneType' object has no attribute 'label_map'..
please can somebody help me?
the label map "name of class to classify/detect" is missing. maybe in the csv exported from roboflow you have to organise as explained in the minute (2:42) because roboflow export it with a different format.
Great video!
why is it installing TFLITE MODEL MAKER in colab taking forever?
Model maker lite package not installing now as previous now it takes more space tell me solution
How to convert to csv from labelImg?
How to add multiple tflite model on one app
I've changed the number of detections in the efficientnet model (tflite_max_detections) before I train and also in Android Studio with (.setMaxResults) in the ObjectBuilder method but still the maximum number of detections is stuck at 25.
Anyone any solutions?
at 8:51 how did u get a bitmap???
😢.. this was done with android studio..
I am labeling images with LabelImg and with every picture i have, i get 1 xml file. How do i put all those XML files in just one CSV to load it into my model?? im kinda confused here :x
And in the CSV file i have to put the Train, Test or validation?
Thanks
Well, you need not create a CSV for it. It supports PascalVOC too. However, I'm facing one error. While loading the data, an error occurs that says, "Image format not JPEG", despite the images being of the said format. Try your luck and let me know if you're successful.
@@nitintiwari7004 I tried to put all files on Google Drive, and It is succeeded without problem. If uploading to Google drive is not bother you, it is good way.
@@ningbo2197 Thanks for the suggestion. However, I resolved the issue. I pre-processed all the images by programmatically converting them to RGB using PIL library before feeding to the model and it worked just perfect.
@@nitintiwari7004 can you please share the code ? I have tried to feed my data from the drive and it is generating some random xxxx.jpeg path and gives error that file does not exist. I have only 225 images..
@@nitintiwari7004 How do youn create PascalVOC
Please someone teach to me, how to stop training on the way to avoid overfitting?
I cancel object_detector.create() by pushing run button, then "model" variable is not recognized on model.evaluate(test_data).
the validation_data is used for that. when running the object_detector.create() you should pass in validation_data=validation_data.
انا استفدت ايه
Instead of making a video on model training & deploying app, you should have just concentrated on how to train a tflite model. This video is jack of all and master of none. Please make a noob-friendly tutorial for a noob like me who wants to learn TF
None of the links in the description working..
Great stuff 👏 👍