Jon Barker

For 1024x1024 images and target objects around 192x192 you probably don't need to adjust the stride initially. DetectNet with default settings should be sensitive to objects in the range 50-400px....

@fchouteau Set `image_size_x: 512 image_size_y: 512 crop_bboxes: false` in `name: "train_transform"`, i.e. the `type: "DetectNetTransformation"` layer applied at training time only. Everywhere else leave the image size as 1024x1024. That...

@JVR32 DetectNet is not designed to work with datasets of the kind that you describe. A dataset for DetectNet should be images where the object you wish detect is some...

@JVR32 You can annotate bounding boxes on the images in their original size - this is probably desirable so that you can use them in that form in the future....

@JVR32 > I think the bounding boxes don't have to be square?! Correct > Or do I have to set a bigger padding (e.g. 1024 x 1024) and set resize...

@aprentis Can you hover over the graph so that we can the actual numeric results for the metric - it matters greatly whether those numbers are just small or exactly...

Running with --debug I get a lot of output like this: GPU: Activating program (id=2) GPU: Active texture is 0 GPU: Activate texture GPU: Activating buffer (id=9) GPU: Activating buffer...

Deleting `/megatron/fused_kernels/build` is recommended if you have upgraded CUDA versions or moved to different hardware. Those changes will not automatically be detected causing a rebuild of the kernels, which may...

Hi. Thanks for creating this PR. We (NVIDIA) are actually planning to release VLM training functionality in Megatron core in the next couple of weeks. As you may have seen,...

This issue will be addressed in the next few days by an update to `preprocess_data.py` that allows processing a large dataset in multiple partitions and thereby avoiding OOM errors. I'll...