ML_Decoder

ML_Decoder copied to clipboard

Hi, thanks for your great work. The performance of ResNet101 based ML-Decoder on the MS-COCO is impressive. The training details in your paper show that the ResNet101 is pretrained on...

jasonseu

infer.py doesn't work, variable referenced before assignment

4

Hi, thanks for the work! But I think there is an obvious bug at line 56 in src_files/models/utils/factory.py: `model.load_state_dict(state[key], strict=True)` The variable `key` in else branch is used without definition....

PT0X0E

Hello,thank you very much for sharing the work. And I find all the results in paper are under imagenet 21k pretrained model. Considering that a lot of work before used...

THUeeY

Hello, I used the pre-training model of TResnet_M, the image input size is 224, the learning rate is 2e-4 in the paper, the rest of the parameters are consistent with...

limengyang1992

some questions of word2vec extraction

1

I want to know which word2vec model you use in the NUS_WIDE ZSL classification setting. I try to extract word representation by myself, and I use 'glove-wiki-gigaword-300.model' model. But I...

Neko9810

Great Project. Can be ML_decoder applied to Efficientnet models for single label classification? Thanks

osivaz61

Nus-wide data set for zsl giving good mAP value but f1 score is less

1

Hi, thanks for your continuous support, for zero-shot learning with the Nus-Wide data set I got mAp values similar to what you mentioned in your papers but f1-score values are...

aliman80

Hi, the code for A2 configuration training in [Single-label Training Code ] does not run successfully， which is: ``` python -u -m torch.distributed.launch --nproc_per_node=8 \ --nnodes=1 \ --node_rank=0 \ ./train.py...

Kyrie10favor

I have a problem with increasing the accuracy or decreasing the loss when using the ml decoder.

5

Thanks for the implementation and great project! I actually saw a similar issue to mine, which was solved by decreasing the learning rate. I tried the same, but there seems...

penguin1109

https://github.com/Alibaba-MIIL/ML_Decoder/blob/8a9e984f671c9c30c98d2c45dfcaf4383381c254/src_files/ml_decoder/ml_decoder.py#L60 Thanks for sharing your work! Why is the dropout layer positioned here? What's the effect of it? Is it noise for generalization?

developer0hye