Problem when use best_model_pascal_ResFCN.pth

[ThanhNguyenFG]

Hello,
After I ran the command

python main.py -image_path figures/test.png
-model_path checkpoints/best_model_pascal_ResFCN.pth
-model_name ResFCN

I got this error:

"Model: ResFCN - Dataset: pascal - Metric: mRMSE
Traceback (most recent call last):
File "main.py", line 45, in
main()
File "main.py", line 33, in main
applyOnImage.apply(args.image_path, args.model_name, args.model_path)
File "/content/LCFCN/applyOnImage.py", line 34, in apply
imsave(save_path, ut.combine_image_blobs(image_raw, pred_blobs))
File "/usr/local/lib/python3.6/dist-packages/skimage/io/_io.py", line 144, in imsave
return call_plugin('imsave', fname, arr, plugin=plugin, **plugin_args)
File "/usr/local/lib/python3.6/dist-packages/skimage/io/manage_plugins.py", line 210, in call_plugin
return func(*args, **kwargs)
File "/usr/local/lib/python3.6/dist-packages/imageio/core/functions.py", line 253, in imwrite
raise ValueError("Image must be 2D (grayscale, RGB, or RGBA).")
ValueError: Image must be 2D (grayscale, RGB, or RGBA)."

I almost try many images but I can't fix this error.
Did I miss something?
Thank you very much.

[sirtris]

I had the same error. I think is because the pred_blobs are of shape [num_classes, img_width, img_height] and the ´image_raw´ is of shape [img_width, img_height, 3] so there is a shape mismatch.
What I did was: pred_blobs_max = np.argmax(pred_blobs, axis=0) and the replace pred_blobs with pred_blobs_max in the imsave line.

However my predictions are very bad I am not sure if I made a mistake here or somewhere else.
Does this solution work for you?

[ThanhNguyenFG]

Thank you very much,
I fixed and no longer have this error, but the result is exactly what you said (it is very bad) and predict_count is always equal to 0 although blobs has appeared on the image.
.

[tongpinmo]

@ThanhNguyenFG @sirtris
I have also come accross the problem, according to your method, I fixed my problem, but the result is so strange, my predict_count is normal but the coordinates is on the top of images

[datkungfu]

@ThanhNguyenFG @sirtris @tongpinmo I have found out the problem with this code. In the file applyOnImage.py you can see the line "pred_counts = int(model.predict(batch, method="counts").ravel()[0])"
But actually, it should become pre_counts= model.predict(batch, method="counts").ravel()
Then, you can add "print(pre_counts)" right after that line and then when we run the file main.py , the result of "print(pre_counts)" would be an array of 20 numbers, they are the number of objects for each class in 20 classes:
"aeroplane":0,
"bicycle":1,
"bird":2,
"boat":3,
"bottle":4,
"bus":5,
"car":6,
"cat":7,
"chair":8,
"cow":9,
"diningtable":10,
"dog":11,
"horse":12,
"motorbike":13,
"person":14,
"pottedplant":15,
"sheep":16,
"sofa":17,
"train":18,
"tvmonitor":19
In the code, author just count the number of object of the first class, "aeroplane".
Next, when we use pred_blob_max as @sirtris , then it would create the density map for the class with highest number of object.