soldierofhell
soldierofhell
Also got this error. I noticed that on my rosbag it appears when robot starts to move. I thought that starting from stationary state is the preferred way to use...
The error is triggered by this [smoother_->update()](https://github.com/MIT-SPARK/Kimera-VIO/blob/641576fd86bdecbd663b4db3cb068f49502f3a2c/src/backend/VioBackend.cpp#L1212-L1213) and originates in gtsam [Values.cpp](https://github.com/borglab/gtsam/blob/cd3854a1f6db923d40ecf3ced56bafbe339d1b3c/gtsam/nonlinear/Values.cpp#L224-L230) when trying to access value by key [Values-inl.h](https://github.com/borglab/gtsam/blob/39b1b15c3e4ffaea1aab4e764b2bcb3a9baa5373/gtsam/nonlinear/Values-inl.h#L345-L357). In VioBackend new_values_ are inserted [here](https://github.com/MIT-SPARK/Kimera-VIO/blob/641576fd86bdecbd663b4db3cb068f49502f3a2c/src/backend/VioBackend.cpp#L740-L743) and cleared [here](https://github.com/MIT-SPARK/Kimera-VIO/blob/641576fd86bdecbd663b4db3cb068f49502f3a2c/src/backend/VioBackend.cpp#L1038). Maybe...
Here is an [issue](https://github.com/borglab/gtsam/issues/301) on gtsam repo also with smart factors, but might not be related
Hi @ebrevdo, There's short note in the code: https://github.com/tensorflow/agents/blob/3448c9e88fbe48d515c85ffab2b96e9f429a3b7d/tf_agents/policies/policy_saver_test.py#L358-L359 I guess this "native support for RNG ops, atan, etc." relates to unsupported BroadcastArgs and BroadcastTo ops. Could you please provide...
Thanks guys, please leave here a comment when BroadcastArgs will be available
@thaink any ETA for this BroadcastArgs issue? :)
> #### Mask Loss: > ``` > mask_loss = 5 * l1_loss(predict_tensor, target_tensor) > predict_tensor = Tensor[B=positive_bbox_num, C={bg, fg}, H=28, W=28] > ``` > > note: `bg=background`, `fg=foreground` > `target_tensor`'s...
I found I had also PMTD mask predictor "MaskRCNNC4Predictor_Upsample". When I switched back to "MaskRCNNC4Predictor" mask loss seems alright now :) Any idea why bilinear upsampling causes trouble?
@JingChaoLiu, thank you for pointing this out to me. Indeed there's `sigmoid()` in your `MaskRCNNC4Predictor_Upsample` head and at this moment this is the cause of bad convergence (I mean if...
My plan for now is to: * inspect gradients in `sigmoid()` version * train longer non-`sigmoid()` version * freeze batch norm (?) * try diffrent loss like mse Any other...