Sharat Agarwal comments

Results 12 comments of


                                            Sharat Agarwal

Expected tensor for argument #1 'input' to have the same type as tensor for argument #2 'rois'; but type torch.cuda.HalfTensor does not equal torch.cuda.FloatTensor

I am also getting the same error. Is the error fixed?

Warm up

Hi @RogerZhangzz , could you please provide the warmup weights for Synthia-> Cityscapes

How to visualize the learned offsets?

> The visualizations require a forward run of the object detection model on the MS-COCO validation dataset. Maybe I will release the visualization codes after the release of the object...

Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2)

Yes please update the results.

Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2)

For the warmup you have used AdaptNet which uses DeeplabV2 as the Segmentation model, was that also changed to DeeplabV3+? If yes can you please report the initial result after...

Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2)

> > For the warmup you have used AdaptNet which uses DeeplabV2 as the Segmentation model, was that also changed to DeeplabV3+? If yes can you please report the initial...

Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2)

> Sorry for the delay. @zhijiew @sharat29ag The result on DeeplabV2 is implemented. Beyond expectation, the final result is 65.16 (mIoU), better than 64.89 (mIoU) with DeeplabV3+. Here is the...

Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2)

> Thanks.

Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2)

@munanning does the warmup weights at sgate1 for GTA->City and Synthia->City were same?

I use my dataset, when run the diversity_sampler, it encounter "assert len(cluster_i) >= 1" problem

@ChenhongyiYang I also faced this error while training on COCO.