Aboud black mask image when process local custom data

[MaybeShewill-CV]

Lots of issues mentioned about the black mask image problem during process their own custorm data. The problem is mainly caused by the cluster postprocess strategy. Here is a sample collected in the closed issues(#382). User have found black mask image when using pretrained model to inference his own image data. The test results are listed here.
mask_image_before_adjust

You may find the instance segmentation result may somehow identify different lanes although it was not that good here but it's a another question. Why would the mask image still become empty? The reason was the dbscan cluster params was not proper configured for that data. When I enlarge the dbscan eps param from 0.35 to 0.5 and reduce min_samples params from 1000 to 250. The cluster could work well on this data. The test was listed here.
mask_image_after_adjust

So first thing you may need to do when faced with such a problem is to adjust the dbscan params to see if it helps. But if you can not even identify different lanes on instance segmentation result or binary segmentation result what you need is to retrain the model rather than adjust some of the params.

Anyway hope it helps you guys !!!

[arijit-imerit]

Hi, I have downloaded the contents of the Dropbox link and extracted it in a directory named weights/
After that I run this command: python3 test_lanenet.py --weights_path weights/ --image_path ~/Downloads/sample.jpg

I am getting the following error:

File "/home/imerit/.local/lib/python3.8/site-packages/tensorflow/python/training/saver.py", line 1290, in restore
raise ValueError("The passed save_path is not a valid checkpoint: " +
ValueError: The passed save_path is not a valid checkpoint: weights/

Can you kindly help me out?

[MaybeShewill-CV]

@arijit-imerit --weights ./weights/xxxx.ckpt

[arijit-imerit]

Thank you for replying!
Unfortunately, the Dropbox link mentioned in the ReadME doesn't contain a .ckpt file. There's a checkpoint, a.meta file, a .index file. The other link for BaiduNetDisk isn't working. Would you kindly be able to upload the .ckpt file in the Dropbox?

[yxgi5]

in file tools/test_lanenet.py

add

import sys
sys.path.append('/ext_drv/mov/lanenet-lane-detection')   # 

change

    with sess.as_default():
        saver.restore(sess=sess, save_path=weights_path)

to

    with sess.as_default():
        module_file = tf.compat.v1.train.get_checkpoint_state(weights_path)
        saver.restore(sess, module_file.model_checkpoint_path)

change checkpoint file

model_checkpoint_path: "tusimple_lanenet.ckpt"
all_model_checkpoint_paths: "tusimple_lanenet.ckpt"

run test like

python tools/test_lanenet.py --weights_path ./weights/tusimple_lanenet/ --image_path ./data/tusimple_test_image/0.jpg

[MaybeShewill-CV]

@arijit-imerit You may google how to restore tensorflow model from checkpoint file:)