Some issues about Chest X-ray classification

We would like to give it a try with Chest X-ray classification, but some issues are occurred. Hope somebody could shed some light on them.

  1. The official manual indicates dataset of Chest X-ray classification is provided by https://biometry.nci.nih.gov/cdas/learn/plco/images/. It is difficult for users who are outside the US to obtain. So we try to download another PLCO dataset from Kaggle and it provides 1024x1024 png files. What are the next steps if we wanna train with Chest X-ray classification model directly to make sure it indeed works?

  2. When downloading classification_chest_xray model within Clara Train SDK container, one error message is visible. After the verification in the target folder, classification_chest_xray.json is indeed unavailable. Meanwhile, the model is unavailable via http://localhost/v1/models .

[2019-08-13 11:23:49.386] INFO - Check for model+config at: /var/nvidia/aiaa/downloads/host-5000/classification_chest_xray/classification_chestxray_v1
[2019-08-13 11:23:49.386]ERROR - Invalid Model Config: /opt/nvidia/medical/tlt2/src/apps/aas/actions/…/configs/models/classification_chest_xray.json
[2019-08-13 11:23:49.386]ERROR - Check if Model Package has config_aiaa.json in {ROOT}, {ROOT}/config

  1. medical/segmentation/examples/brats/tutorial_brats.ipynb and /opt/nvidia/medical/classification/examples/PLCO mentioned in the official Clara manual are unavailable within Clara Train SDK container. Where can we get them?

Highly appreciated for any feedback.

1 Like