Training for 80000 times

[jiesico]

Hi! Thank you very much for sharing the code of DROID_SLAM. I've trained 80,000 times on the Tartanair dataset so far, but why hasn't loss_function shown any signs of convergence so far?

[fabiopoiesi]

Hi,

We are experiencing a loss like this during training Tartanair.

Is this something expected?

Thanks

[xhangHU]

Can you use the weight file you got for testing? When I test with the weights I got from training, the following error is reported: Traceback (most recent call last): File "/root/docker2/droid/2new/DROID-SLAM/demo.py", line 117, in traj_est = droid.terminate(image_stream(args.imagedir, args.calib, args.stride)) File "/root/docker2/droid/2new/DROID-SLAM/droid_slam/droid.py", line 81, in terminate self.backend(7) File "/root/anaconda3/envs/droidenv5/lib/python3.9/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context return func(*args, **kwargs) File "/root/docker2/droid/2new/DROID-SLAM/droid_slam/droid_backend.py", line 33, in call graph.add_proximity_factors(rad=self.backend_radius, File "/root/docker2/droid/2new/DROID-SLAM/droid_slam/factor_graph.py", line 368, in add_proximity_factors ii, jj = torch.as_tensor(es, device=self.device).unbind(dim=-1) ValueError: not enough values to unpack (expected 2, got 0)

[realXiaohan]

@xhangHU Hi, I have run the model with the weights they provide successfully but met the same problem as I used my self-trained weights. Did u solve this problem yet?

[xhangHU]

@xhangHU Hi, I have run the model with the weights they provide successfully but met the same problem as I used my self-trained weights. Did u solve this problem yet?

It's not solved yet, and I checked that the network structure used in the pre-training model provided by the author is not the same as in the code provided

[lvmingzhe]

@xhangHU Hi, can I have your email please? We can talk about it in more detail.

Hi, could I join yours? I would like to retrain the model but got stuck in the beginning. Thanks a lot. 12131040[at]mail.sustech.edu.cn

[519174419]

@xhangHU Hi, I have run the model with the weights they provide successfully but met the same problem as I used my self-trained weights. Did u solve this problem yet?

It's not solved yet, and I checked that the network structure used in the pre-training model provided by the author is not the same as in the code provided

It maybe cause by, the distance of graph is loss than the thresh. Try to give a smaller thresh, the demo will run successfully. But I still meet some problem with the disps result

[realXiaohan]

@519174419 Yeah, I solved the problem by setting a smaller thresh but the ego-motion prediction looks so weird with my self-trained model. Is anything wrong with your disparity map?

[519174419]

**realXiaohan ** commented 16分钟前

Yes,I think the flow_loss and geo_loss is low, but it still something wrong with the disparity map.

[realXiaohan]

@519174419 We can talk about it in more detail and my email is [[email protected]].

[YznMur]

@519174419 Yeah, I solved the problem by setting a smaller thresh but the ego-motion prediction looks so weird with my self-trained model. Is anything wrong with your disparity map?

Hi @realXiaohan , could you please share your config for training and demo ?