mAP training several classes = 0.0 and so low with data custom using detectnet_v2 (resnet_18))

I use .hdf5 as pretrain in my config, with latest docker tf1 output return .hdf5 checkpoint, I wanna output return .tlt

The .tlt is encrypted .hdf5 file. You can directly run inference or evaluation with the .hdf5 file. The .tlt file is not needed for TAO5.0. Please refer to tao_tutorials/notebooks/tao_launcher_starter_kit/detectnet_v2/specs/detectnet_v2_inference_kitti_tlt.txt at main · NVIDIA/tao_tutorials · GitHub and tao_tutorials/notebooks/tao_launcher_starter_kit/detectnet_v2/detectnet_v2.ipynb at main · NVIDIA/tao_tutorials · GitHub.

mAP at epoch 1, 2 is 0.0 .I think with large labels with each class. mAP is ~ 0.5 at least. And I set enable_autoweighting: false. What does this param mean?

To narrow down, please train only person class instead. You can modify the training spec file and trigger training.

should i set enable_autoweighting: false? And should i resize images training offline or set enable_auto_resize=true?

Set enable_autoweighting: true and enable_auto_resize=true.

log_train_resized.txt (122.9 KB)
mAP=0.0
this my config_train:
config_train_resized.txt (10.0 KB)

I resized images training offline, but my loss and mAP so bad

Can you help me? Please!
log_training_terminal.txt (2.7 MB)

Firstly, please double check the groundtruth are correct.
And you need not resize images offline when you set enable_auto_resize=true.

If possible, can you share one training image and its label file? Thanks.

I have checked my groundtruth
Ex: xmin, ymin, xmax, ymax

car 0.00 0 0.00 587.01 173.33 614.12 200.12 0.00 0.00 0.00 0.00 0.00 0.00 0.00

I visualized objects, and it correct.

So, seems that the objects are small. To narrow down, please try to train person class only, also modify as mentioned in mAP training several classes = 0.0 and so low with data custom using detectnet_v2 (resnet_18)) - #20 by Morganh.

I tried to train person class only, also modify as mentioned above, and i check object dimention, It isn’t small, but mAP =0. I’ll try inference and no response(no object detected).

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

Please try to train with attached training spec file.
20240201_280630.txt (3.7 KB)

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.