zhengpeng7 / w-net-keras

An unofficial implementation of W-Net for crowd counting.

W-Net-Keras

Unofficial implementation of W-Net for crowd counting in Keras.

---

Paper:

• Original_paper: Varun Kannadi Valloli, Kinal Mehta. W-Net: Reinforced U-Net for Density Map Estimation.

Results now:

On dataset ShanghaiTech B

Under development...

MAE	MSE	Mean of Frobenius Norm	MAPE	PSNR	SSIM
7.85	12.17	6.75e-7	6.46%	29.27	0.93

Dataset:

• ShanghaiTech dataset: dropbox or Baidu Disk.

Training Parameters:

1. Loss = MSE * 1000 + BCE * 10;

2. Optimizer = Adam(lr=1e-4, decay=5e-3);

3. Batch size: 1;

4. Data augmentation: Flip horizontally online randomly;

5. Patch: No patch;

6. Batch normalization: No BN layers at present;

7. Weights: Got best weights in epoch248(250 epochs in total), and here is the loss records:

   

8. Prediction example:

   

Run:

1. Download dataset;
2. Data generation: run thegenerate_datasets.ipynb .
3. Run the main.ipynb to train, test, analyze and evaluate the image quality.

w-net-keras's Issues

Nice work！Is this part for loading pretrained VGG model?And it is a way to initialize the layer's weight?

front_end = VGG16(weights='imagenet', include_top=False)
weights_front_end = []
for layer in front_end.layers:
if 'conv' in layer.name:
weights_front_end.append(layer.get_weights())
counter_conv = 0
for i in range(len(model.layers)):
if counter_conv >= 13:
break
if 'conv' in model.layers[i].name:
model.layers[i].set_weights(weights_front_end[counter_conv])
counter_conv += 1

Release of the pretrained model on SHB

Could you share the pretrained weights of W-Net on SHB.

When generating the dataset an Error occurs: ValueError: operands could not be broadcast together with shapes (568,15) (0,15) (568,15)

data/ShanghaiTech/part_A/train_data\images\IMG_118.jpg
4%|▍ | 21/482 [00:02<00:45, 10.16it/s]
Traceback (most recent call last):
File "C:/Users/a/PycharmProjects/W-Net-Keras/generate_datasets.py", line 46, in
DM = gen_density_map_gaussian(k, gt, sigma=sigma)
File "C:\Users\a\PycharmProjects\W-Net-Keras\utils_gen.py", line 86, in gen_density_map_gaussian
] += gaussian_map
ValueError: operands could not be broadcast together with shapes (568,15) (0,15) (568,15)

The same problem occurs in your other code, SANet-Keras. I'm using Python 3.7.

Problem for creating classification map

From original paper"we used the same method as density generation but
used a larger window size and spread parameter. Once the
blurred map is generated, we use binary thresholding to
create classification map to train the reinforcement branch.
In our experiments, we used a threshold (th) = 0.001 ",I can't find the Corresponding content in your code ,will you help me? Please tell me the specific location in your code.

Dataset isn't accessibly

I can't download dataset from dropbox neither baidu.