ashishpatel26/skin-lesions-detection-deep-learning

Transfer Learning with DCNNs (DenseNet, Inception V3, Inception-ResNet V2, VGG16) for skin lesions classification on HAM10000 dataset largescale data.

Skin Lesions Classification with Deep Convolutional Neural Network

Dataset

---

The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions

---

Files Tables

Sr.No	File Name	Link
1	Exploratory data analysis	Notebook
2	Baseline model	Notebook
3	Fine-tuning the last convolutional block of VGG16	Notebook
4	Fine-tuning the top 2 inception blocks of InceptionV3	Notebook
5	Fine-tuning the Inception-ResNet-C of Inception-ResNet V2	Notebook
6	Fine-tuning the last dense block of DenseNet 201	Notebook
7	Fine-tuning all layers of pretrained Inception V3 on ImageNet	Notebook
8	Fine-tuning all layers of pretrained DenseNet 201 on ImageNet	Notebook
9	Ensemble model of the fully fine-tuned Inception V3 and DenseNet 201	Notebook

Technical Issue

All Notebooks using Keras 2.2.4 and Tensorflow 1.11. Batch-Norm layer in this version of Keras is implemented in a way that: during training your network will always use the mini-batch statistics either the BN layer is frozen or not; also during inference you will use the previously learned statistics of the frozen BN layers. As a result, if you fine-tune the top layers, their weights will be adjusted to the mean/variance of the new dataset. Nevertheless, during inference they will receive data which are scaled differently because the mean/variance of the original dataset will be used. Consequently, if use Keras's example codes for fine-tuning Inception V3 or any network with batch norm layer, the results will be very bad. Please refer to issue #9965 and #9214. One temporary solution is:

for layer in pre_trained_model.layers:
    if hasattr(layer, 'moving_mean') and hasattr(layer, 'moving_variance'):
        layer.trainable = True
        K.eval(K.update(layer.moving_mean, K.zeros_like(layer.moving_mean)))
        K.eval(K.update(layer.moving_variance, K.zeros_like(layer.moving_variance)))
    else:
        layer.trainable = False

Results

Models	Validation	Test	Depth	# Params
Baseline	77.48%	76.54%	11 layers	2,124,839
Fine-tuned VGG16 (from last block)	79.82%	79.64%	23 layers	14,980,935
Fine-tuned Inception V3 (from the last 2 inception blocks)	79.935%	79.94%	315 layers	22,855,463
Fine-tuned Inception-ResNet V2 (from the Inception-ResNet-C)	80.82%	82.53%	784 layers	55,127,271
Fine-tuned DenseNet 201 (from the last dense block)	85.8%	83.9%	711 layers	19,309,127
Fine-tuned Inception V3 (all layers)	86.92%	86.826%	_	_
Fine-tuned DenseNet 201 (all layers)	86.696%	87.725%	_	_
Ensemble of fully-fine-tuned Inception V3 and DenseNet 201	88.8%	88.52%	_	_

The Dataset

The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions

skin-lesions-detection-deep-learning's Issues

Nice work and need informations

Hi,
Just to say thank you because you make some great work, I mean your results seem honest compared to some studies that achieved something like 99% without providing any information. And you put your sources, the dataset, the faced issues with batch norm (I got the same trouble).
I planned to reuse your DenseNet approach on my dermoscopy dataset, and that's my point : did you have any published work on these results? I would like to cite your work on a paper I'm currently writting?
Have a nice day

how can i get the csv dataset

Any reference on how to get the csv dataset. I didn't find it on the given weblink. Help would be appreciated.
Thanks

checkpoint

Hello and thank you for sharing your work!
Would it be possible for you to provide the model checkpoint for inference?
Thank you in advance, Lucia