Abhishek Agrawal

Deploy non-mmlab model using mmdeploy

1

How can I deploy a model like [YOLACT++](https://github.com/dbolya/yolact) using mmdeploy ?

Q1. Can I train a model from scratch using Quantization aware training? Or It can only be used as fine-tuning method ? Q2. Are QFormats for weight tensors decided before...

Changed related to ONNX support for yolact++ added

1

Hey, I have successfully added support for Onnx export and inference for yolact++ model. I would like the collaborators to have a look at the PR, access and merge it...

SWIN transformer pretrained model issue

2

## About this repo - commit - https://github.com/wang-xinyu/tensorrtx/commit/0f944b89750c2cd2e6be7933f84d61eb4e34ec83? - model - SWIN-Transformer ## Your problem ``` python gen_wts.py upernet_swin_tiny_patch4_window7_512x512.pth Traceback (most recent call last): File "gen_wts.py", line 8, in model...

I read the concept of **Matching predictions to ground truths**. You mentioned to check jaccard overlap for the matching. Did you follow any other matching criterion?

Different Accuracy with different batch-size

1

I am getting significant difference in the validation accuracy when I am using different batch size. Can anyone please explain this ?

RuntimeError: a view of a leaf Variable that requires grad is being used in an in-place operation.

2

While working with torch=1.8.0 I was seeing this error. The error is because of the following line of code. https://github.com/WongKinYiu/PyTorch_YOLOv4/blob/master/models/models.py#L203-L210 Change this to - j = layers[yolo_index] if 'from' in...

Export to ONNX

3

hey @lbin @CharlesShang, Thanks for the work. Can you please add a script which is able to export _DCN_v2 to ONNX and run inference with onnxruntime ?

Can you guide me about, which parameter should be tracked? And how do tune the measurement and process noise for this type of application. Here I am getting detection from...

Evaluation is too slow

5

Hey @mimoralea As the evaluation is being sequentially for each image, the process is too slow. Is there any other way for evaluation?