ZhaochongAn
vis
Hello, I saw the visualization comparison results between your model and other models in your paper. However, I have read many previous papers that did not provide visualization results as you did to visually see the prediction effect. May I ask how you visualized the network model effect of the previous comparison experiment? Do you have any good methods or code? Because the previous code was not provided, I have been troubled for a long time.
Hi, in our experiments, we only compare to the previous methods that release the code. Based on their code, we retrain the model in our setting and use our visualization code to get the results. Our visualization code is at util/visualize.py (https://github.com/ZhaochongAn/COSeg/blob/main/util/visualize.py) and you can find the instructions on visualization at our README page.
Hello, I have successfully visualized my own dataset. Thank you for your reply. As other public networks currently do not have visualization code to save prediction files for visualization, you just replied that training should be done in your settings. I would like to ask specifically where I can modify your network to evaluate the results of other networks in your settings?
For other networks you want to evaluate, you could write their model file by following the input and output format in the forward function of coseg.py (https://github.com/ZhaochongAn/COSeg/blob/main/model/coseg.py). Then, you could train it and visualize the results using our provided code.
I noticed that you used DGcnn as a pre training backbone when evaluating other models. So, did you use it to modify the input-output format in your own CoSeg model after obtaining weights from pre training other models, and then visualize it with fewer samples? Because I found it difficult to replace your pre training backbone with DGcnn for pre training before. That is to say, is it possible to use your framework to pre train dgcnn and then modify CoSeg.by for small sample training?
---Original--- From: "Zhaochong @.> Date: Mon, Sep 30, 2024 18:36 PM To: @.>; Cc: @.@.>; Subject: Re: [ZhaochongAn/COSeg] vis (Issue #13)
For other networks you want to evaluate, you could write their model file by following the input and output format in the forward function of coseg.py (https://github.com/ZhaochongAn/COSeg/blob/main/model/coseg.py). Then, you could train it and visualize the results using our provided code.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
I noticed that you used DGcnn as a pre training backbone when evaluating other models. I also tried to replace your pre training backbone with DGcnn for pre training and found it difficult. So, are you conducting pre training on the original model to obtain model weights, and then modifying the input and output format of CoSeg.Py's forward function in your model for few shot training
---Original--- From: "Zhaochong @.> Date: Mon, Sep 30, 2024 18:36 PM To: @.>; Cc: @.@.>; Subject: Re: [ZhaochongAn/COSeg] vis (Issue #13)
For other networks you want to evaluate, you could write their model file by following the input and output format in the forward function of coseg.py (https://github.com/ZhaochongAn/COSeg/blob/main/model/coseg.py). Then, you could train it and visualize the results using our provided code.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
Hey, for the dgcnn exps, we are conducting the pretraining and episode training using this codebase by adding a dgcnn backbone file. Other parts are kept same. So, you just need to make sure the output is kept as the input format for COSeg and no need to change the COSeg model.