Dheeraj R Reddy
Dheeraj R Reddy
@dumblob Thanks for creating this ticket. Your rationale makes sense, I'll start adding comparison benchmarks soon. I've found a couple of other libraries that might be good reference points for...
@jian-li Can you share the code that resulted in the memory leak? That would help me with debugging.
Here's a [python](https://github.com/DHZS/tf-deformable-conv-layer/blob/master/nets/deformable_conv_layer.py) only implementation of v1 for Keras which can be ported to TF-Keras. [This repository](https://github.com/msracver/Deformable-ConvNets) is the official implementation with CUDA code; it's written using MXNet but it...
@rootkitchao Any update? I can work on this if you aren't.
Using native CUDA for deform conv is dependant on #118 .
@vlordier Nope, haven't found the time to get around to this. Feel free to open a PR if you decide to work on this.
@gabrieldemarmiesse Sorry about the delay. Moving [`average_wrapper`](https://github.com/tensorflow/addons/blob/master/tensorflow_addons/optimizers/average_wrapper.py) would be fairly easy. We need to override `transform_unaggregated_gradients`only. Once `average_wrapper` is migrated, migrating `moving_average` and `stochastic_weight_average` optimizers would be very simple.
Here's an idea to get started: Tracking idle threads. Before a thread goes to sleep (`sem.acquire_many`), add it to an idle thread queue. Whenever a task needs to be enqueued,...
Of course. Time to RIIR.