Support for ROI Pooling #592

varunagrawal · 2018-08-30T19:05:34Z

ROI Pooling with tests. Fix for cuda context in ROI Align.

@fmassa please can you review this?

fmassa

It looks very nice, thanks!

I particularly liked that you added support for non-contiguous grad_output, that was a bug in the previous implementation.

Could you address the few comments I've added? Otherwise looks good to merge!

torchvision/csrc/cuda/ROIPool_cuda.cu

@@ -193,7 +197,11 @@ at::Tensor ROIPool_backward_cuda(const at::Tensor& grad,
         pooled_height,
         pooled_width,
         grad_input.data<scalar_t>(),
-         rois.data<scalar_t>());
+         rois.data<scalar_t>(),
+         n_stride,


torchvision/csrc/cuda/ROIPool_cuda.cu

-          offset_bottom_diff + argmax,
-          static_cast<T>(offset_top_diff[ph * pooled_width + pw]));
-
+    const int pooled_height, const int pooled_width, T* bottom_data,


torchvision/csrc/cuda/ROIPool_cuda.cu

@@ -30,7 +28,7 @@ __global__ void RoIPoolFForward(const int nthreads, const T* bottom_data,
    int roi_end_w = round(offset_bottom_rois[3] * spatial_scale);
    int roi_end_h = round(offset_bottom_rois[4] * spatial_scale);

-    // Force malformed ROIs to be 1x1
+    // Force malformed ROIs to be 1x1 or HxW


torchvision/csrc/cuda/ROIPool_cuda.cu

+    const int n_stride, const int c_stride,
+    const int h_stride, const int w_stride) {
+
+    CUDA_1D_KERNEL_LOOP(index, nthreads) {


fmassa

Thanks a lot!

This reverts commit ede93b2.

andfoy · 2018-10-12T01:48:12Z

@fmassa @varunagrawal I'm unable to compile the latest revision of layers branch with the most recent PyTorch nightly wheels, it fails with the following error traceback:

in file included from /tmp/pip-req-build-2943ca2h/torchvision/csrc/cpu/vision.h:2:0,
                     from /tmp/pip-req-build-2943ca2h/torchvision/csrc/cpu/ROIAlign_cpu.cpp:1:
    /home/eamargffoy/anaconda3/envs/pytorch_041/lib/python3.6/site-packages/torch/lib/include/torch/csrc/api/include/torch/torch.h:13:2: warning: #warning "Including torch/torch.h for C++ extensions is deprecated. Please include torch/extension.h" [-Wcpp]
     #warning "Including torch/torch.h for C++ extensions is deprecated. Please include torch/extension.h"
      ^~~~~~~
    /tmp/pip-req-build-2943ca2h/torchvision/csrc/cpu/ROIAlign_cpu.cpp: In function ‘at::Tensor ROIAlign_forward_cpu(const at::Tensor&, const at::Tensor&, float, int, int, int)’:
    /tmp/pip-req-build-2943ca2h/torchvision/csrc/cpu/ROIAlign_cpu.cpp:234:36: error: ‘struct at::Type’ has no member named ‘tensor’; did you mean ‘renorm’?
       at::Tensor output = input.type().tensor({num_rois, channels, pooled_height, pooled_width});
                                        ^~~~~~
    error: command 'gcc' failed with exit status 1
error
Cleaning up...

Could you please give me some directions about which changes must be done in order to compile this branch without any errors?

Thanks

varunagrawal · 2018-10-12T18:05:54Z

@andfoy thanks for reporting this. This is an issue in ROIAlign and not ROIPool. This PR is specific to ROIPool. I am still working on fixing this. Thank you for your patience.

I'll get back to you on fixes for this really soon.

andfoy · 2018-10-12T19:34:18Z

@varunagrawal Thanks for your reply, is this error related to: pytorch/pytorch#12360?

varunagrawal · 2018-10-12T19:44:13Z

Yep. That is exactly it.

andfoy · 2018-10-12T20:01:57Z

@varunagrawal Do you think that if all tensor creation operations are replaced by the new API, RoIAlign/Pool would work?

fmassa · 2018-10-13T20:04:26Z

Thanks for updating the layers branch @varunagrawal
FYI, we will be releasing in Oct 24th an official set of layers for detection.

varunagrawal · 2018-10-15T17:25:25Z

@fmassa can you please elaborate on the last comment? Will the layers be a part of pytorch proper or a part of torchvision?

More importantly, will this be just a list of approved layers or an actual release of code?

fmassa · 2018-10-16T11:35:06Z

@varunagrawal my comment was not very clear, sorry. I meant we will be releasing an object detection library, which will contain the ROI layers.

The layers will initially be part of this library. They might be moved out into torchvision at a later moment, but not initially

varunagrawal · 2018-10-16T13:48:42Z

Ah gotcha. I did talk to @soumith about this in 2017, so feel free to rope me in.

* ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions

@lopuhin

* Initial layout for layers with cpp extensions * Move files around * Fix import after move * Add support for multiple types to ROIAlign * Different organization CUDA extensions work now * Cleanups * Reduce memory requirements for backwards * Replace runtime_error by AT_ERROR * Add nms test * Add support for compilation using CPP extensions * Change folder structure * Add ROIPool cuda * Cleanups * Add roi_pool.py * Fix lint * Add initial structures folder for bounding boxes * Assertion macros compatible with pytorch master (#540) * Support for ROI Pooling (#592) * ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions * remove .type().tensor() calls in favor of the new approach to tensor initialization (#626) * Consistent naming for rois variable (#627) * remove .type().tensor() calls in favor of the new approach to tensor initialization * Consistent naming for rois variable in ROIPool * ROIPool: Support for all datatypes (#632) * Use of torch7 naming scheme for ROIAlign forward and backward * use common cuda helpers in ROIAlign * use .options() in favor of .type() where applicable * Added tests for forward pass of ROIAlign, as well as more consistent naming scheme for CPU vs CUDA * working ROIAlign cuda backwards pass * working ROIAlign backwards pass for CPU * added relevant headers for ROIAlign backwards * tests for ROIAlign layer * replace .type() with .options() for tensor initialization in ROIAlign layers * support for Half types in ROIAlign * gradcheck tests for ROIAlign * updated ROIPool on CPU to work with all datatypes * updated and cleaned tests for ROI Pooling * Fix rebase problem * Remove structures folder * Improve cleanup and bugfix in test_layers * Update C++ headers * Add CUDAGuard to cu files * Add more checks to layers * Add CUDA NMS and tests * Add multi-type support for NMS CUDA * Avoid using THCudaMalloc * Add clang-format and reformat c++ code * Remove THC includes * Rename layers to ops * Add documentation and rename functions * Improve the documentation a bit * Fix some lint errors * Fix remaining lint inssues * Area computation doesn't add +1 in NMS * Update CI to use PyTorch nightly * Make NMS return indices sorted according to the score * Address reviewer comments * Lint fixes * Improve doc for roi_align and roi_pool * move to xenial * Fix bug pointed by @lopuhin * Fix RoIPool reference implementation in Python 2 Also fixes a bug in the clip_boxes_to_image -- this function needs a test! * Remove change in .travis

ROI Pooling with tests. Fix for cuda context in ROI Align.

7b6d1d2

This was referenced Aug 30, 2018

RoI Pooling Layer & Tests #587

Closed

ROI Pooling Layer #585

Closed

fmassa requested changes Aug 31, 2018

View reviewed changes

renamed bottom and top to follow torch conventions

21957e5

varunagrawal force-pushed the layers branch from 2f2255e to 21957e5 Compare August 31, 2018 20:17

fmassa reviewed Sep 3, 2018

View reviewed changes

torchvision/csrc/cuda/ROIPool_cuda.cu

const int n_stride, const int c_stride,

const int h_stride, const int w_stride) {

CUDA_1D_KERNEL_LOOP(index, nthreads) {

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

fmassa approved these changes Sep 3, 2018

View reviewed changes

fmassa merged commit ede93b2 into pytorch:layers Sep 3, 2018

SelvamArul added a commit to SelvamArul/vision that referenced this pull request Oct 11, 2018

Revert "Support for ROI Pooling (pytorch#592)"

accdb52

This reverts commit ede93b2.

varunagrawal added a commit to varunagrawal/vision that referenced this pull request Jan 24, 2019

Support for ROI Pooling (pytorch#592)

8df8b5e

* ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions

varunagrawal added a commit to varunagrawal/vision that referenced this pull request Jan 25, 2019

Support for ROI Pooling (pytorch#592)

96b6d29

* ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions

varunagrawal mentioned this pull request Mar 11, 2019

[feature request] ROI Pooling layers #477

Closed

3 tasks

fmassa pushed a commit that referenced this pull request Mar 31, 2019

Support for ROI Pooling (#592)

9703ffa

* ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions

fmassa pushed a commit to fmassa/vision-1 that referenced this pull request Apr 15, 2019

Support for ROI Pooling (pytorch#592)

f80d1af

* ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions

fmassa pushed a commit that referenced this pull request May 7, 2019

Support for ROI Pooling (#592)

563604c

* ROI Pooling with tests. Fix for cuda context in ROI Align. * renamed bottom and top to follow torch conventions

rajveerb pushed a commit to rajveerb/vision that referenced this pull request Nov 30, 2023

[minigo] Retire minigo after training v2.1 (pytorch#592)

f098aff

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support for ROI Pooling #592

Support for ROI Pooling #592

varunagrawal commented Aug 30, 2018

fmassa left a comment

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

fmassa left a comment

andfoy commented Oct 12, 2018

varunagrawal commented Oct 12, 2018 •

edited

Loading

andfoy commented Oct 12, 2018

varunagrawal commented Oct 12, 2018

andfoy commented Oct 12, 2018

fmassa commented Oct 13, 2018

varunagrawal commented Oct 15, 2018

fmassa commented Oct 16, 2018

varunagrawal commented Oct 16, 2018

Support for ROI Pooling #592

Support for ROI Pooling #592

Conversation

varunagrawal commented Aug 30, 2018

fmassa left a comment

Choose a reason for hiding this comment

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

fmassa left a comment

Choose a reason for hiding this comment

andfoy commented Oct 12, 2018

varunagrawal commented Oct 12, 2018 • edited Loading

andfoy commented Oct 12, 2018

varunagrawal commented Oct 12, 2018

andfoy commented Oct 12, 2018

fmassa commented Oct 13, 2018

varunagrawal commented Oct 15, 2018

fmassa commented Oct 16, 2018

varunagrawal commented Oct 16, 2018

varunagrawal commented Oct 12, 2018 •

edited

Loading