romroc / objdet_fizyr_colab Goto Github PK

Here are the steps I followed:

• downloaded the pretrained model resnet50_coco_best_v2.1.0.h5
• then I trained it on my dataset using keras_retinanet/bin/train.py csv /path/to/csv/file/containing/annotations /path/to/csv/file/containing/classes
• then I converted the last model saved in snapshots for inference using keras_retinanet/bin/convert_model.py snapshots/resnet50_csv_50.h5 /path/to/save/inference/model.h5

The thing is that I am not getting any bounding boxes on any image. Scores are an array of -1s and the boxes also have negative values. Did I have to mention something explicitly while training the resnet50_coco_best_v2.1.0.h5 on my dataset?
`

Im having trouble getting the relevant CSV's to generate correctly. When running the code, everything seems to work correctly however the labels I have generated are not being passed through to the inference section of the model despite the formatting being correct. On a whim I have tried running the code with the original google drive ID values and dataset and switched back to my own and tested images are still being labelled as goats.

I get this error when I want to run this part

In my gdrive I already use .zip file, but still i get this error.. maybe you have a suggestion to me?

I wanted to train the model for 100 epochs but when the training started after certain epochs the training gets stooped.Why is that happening ??

WARNING:tensorflow:From keras_retinanet/bin/train.py:542: Model.fit_generator (from tensorflow.python.keras.engine.training) is deprecated and will be removed in a future version.
Instructions for updating:
Please use Model.fit, which supports generators.
Epoch 1/100
2020-08-25 04:41:47.091716: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudnn.so.7
2020-08-25 04:41:48.890281: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcublas.so.10
2020-08-25 04:41:49.682925: W tensorflow/core/common_runtime/bfc_allocator.cc:312] Garbage collection: deallocate free memory regions (i.e., allocations) so that we can re-allocate a larger region to avoid OOM due to memory fragmentation. If you see this message frequently, you are running near the threshold of the available device memory and re-allocation may incur great performance overhead. You may try smaller batch sizes to observe the performance impact. Set TF_ENABLE_GPU_GARBAGE_COLLECTION=false if you'd like to disable this feature.
28/50 [===============>..............] - ETA: 6s - loss: 1.7892 - regression_loss: 0.7126 - classification_loss: 1.0766WARNING:tensorflow:Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches (in this case, 5000 batches). You may need to use the repeat() function when building your dataset.

Epoch 00001: saving model to ./snapshots/resnet50_csv_01.h5
28/50 [===============>..............] - 9s 315ms/step - loss: 1.7892 - regression_loss: 0.7126 - classification_loss: 1.0766

Literally, it recognize whatever animal as Goats, i tested it at least three times and i got this results:

Why does it happen?

New URL for 0.5.1 is:

https://github.com/fizyr/keras-retinanet/releases/download/0.5.1/resnet50_coco_best_v2.1.0.h5

Hi RomRoc,

thank you for this script, it helps us a lot transforming annotation from labelImg xml to keras-retinanet csv fizyr/keras-retinanet#908

i have been thinking of doing the other way around, a semi-automatic annotation using a pre-trained model (predicted bounding boxes as annotation), save them to pandas dataframe and convert them to labelImg xml format, therefore we can load the xml and debug/fix/modify the incorrect annotation predicted by the pre-trained model

maybe you could help creating a python script also for this, i cant somehow wrap my head around the logic, greatly appreciate your help

I get the below error when running the line :
!keras_retinanet/bin/train.py --freeze-backbone --random-transform --weights {PRETRAINED_MODEL} --batch-size 8 --steps 500 --epochs 10 csv annotations.csv classes.csv

Traceback (most recent call last):
File "keras_retinanet/bin/train.py", line 547, in
main()
File "keras_retinanet/bin/train.py", line 507, in main
config=args.config
File "keras_retinanet/bin/train.py", line 117, in create_models
model = model_with_weights(backbone_retinanet(num_classes, num_anchors=num_anchors, modifier=modifier, pyramid_levels=pyramid_levels), weights=weights, skip_mismatch=True)
File "keras_retinanet/bin/../../keras_retinanet/models/resnet.py", line 38, in retinanet
return resnet_retinanet(*args, backbone=self.backbone, **kwargs)
File "keras_retinanet/bin/../../keras_retinanet/models/resnet.py", line 99, in resnet_retinanet
resnet = keras_resnet.models.ResNet50(inputs, include_top=False, freeze_bn=True)
File "/usr/local/lib/python3.6/dist-packages/keras_resnet/models/_2d.py", line 188, in ResNet50
return ResNet(inputs, blocks, numerical_names=numerical_names, block=keras_resnet.blocks.bottleneck_2d, include_top=include_top, classes=classes, *args, **kwargs)
File "/usr/local/lib/python3.6/dist-packages/keras_resnet/models/_2d.py", line 66, in ResNet
x = keras_resnet.layers.BatchNormalization(axis=axis, epsilon=1e-5, freeze=freeze_bn, name="bn_conv1")(x)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py", line 926, in call
input_list)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/base_layer.py", line 1117, in _functional_construction_call
outputs = call_fn(cast_inputs, *args, **kwargs)
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/autograph/impl/api.py", line 258, in wrapper
raise e.ag_error_metadata.to_exception(e)
TypeError: in user code:

/usr/local/lib/python3.6/dist-packages/keras_resnet/layers/_batch_normalization.py:17 call  *
    return super(BatchNormalization, self).call(training=(not self.freeze), *args, **kwargs)

TypeError: type object got multiple values for keyword argument 'training'

Any thoughts ?

Hi,

The notebook is really interesting and helpful, however when I try to train the model with your dataset an error pops up as follows.

Creating model, this may take a second...
Traceback (most recent call last):
File "keras_retinanet/bin/train.py", line 514, in
main()
File "keras_retinanet/bin/train.py", line 464, in main
num_classes=train_generator.num_classes(),
File "keras_retinanet/bin/../../keras_retinanet/preprocessing/csv_generator.py", line 170, in num_classes
return max(self.classes.values()) + 1
ValueError: max() arg is an empty sequence

Hope you can help

Cheers
Tarek