• Hardware (T4/V100/Xavier/Nano/etc) Alienware: RTX 4090,
Processor 13th Gen Intel(R) Core™ i9-13900HX (2.20 GHz)
Installed RAM 32.0 GB (31.7 GB usable)
System type 64-bit operating system, x64-based processor
from COURSE: Synthetic Data Generation for Perception Model Training in Isaac Sim https://learn.nvidia.com/courses/course?course_id=course-v1:DLI+S-OV-30+V1&unit=block-v1:DLI+S-OV-30+V1+type@vertical+block@7fecaf9f66204c0ea35402fca5ae1b25
I generated synthetic images using the generate_data.sh script and ran the training model local_train.ipynb.
After the first run it generated the _warning.json file:
"{
"/workspace/tao-experiments/cardbox_sdg/cardbox_data/distractors_warehouse/Camera/object_detection/1106.txt_2": [
471,
393,
471,
428
],
"/workspace/tao-experiments/cardbox_sdg/cardbox_data/distractors_warehouse/Camera/object_detection/476.txt_0": [
721,
239,
721,
264
],
"/workspace/tao-experiments/cardbox_sdg/cardbox_data/distractors_warehouse/Camera/object_detection/1576.txt_2": [
473,
237,
473,
274
]
}"
apparently it created bounding boxes with width = 0.
I’d like to understand why it is doing that and how to avoid it to happen.
I checked my resnet18_distractors.txt spec file and it has a min size for bound boxes already
augmentation_config {
preprocessing {
output_image_width: 960
output_image_height: 544
min_bbox_width: 20.0
min_bbox_height: 20.0
output_image_channel: 3
}