zhangao
zhangao
It seems like the gqa_bal_qla_val.json is actually gqa_bal_qla_testdev.json. The real gqa_all_qla_val.json is merged to gqa_all_qla_train.json.
When applying DataParallel to model.full_im_net, the full_im_net will scatter every param into several part. However, only one param rois need to be divided. You can use "rois_feature = self.full_im_net(full_im.repeat(gpu_num, 1,1,1),...
> /data/anaconda3/envs/scene_graph_benchmark/lib/python3.8/site-packages/torch/include/ATen/core/TensorBody.h:322:1: note: declared here > T * data() const { > ^ ~~ > /data/swfcode/Scene-Graph-Benchmark.pytorch/maskrcnn_benchmark/csrc/cuda/deform_conv_kernel_cuda.cu:855:106: warning: ‘T* at::Tensor::data() const [with T = c10::Half]’ is deprecated [-Wdeprecated-declarations] > const scalar_t...
Yes, you just need to provide the image. The visualization results are of [vg1800_motif_sgcls_I90_E100.pth](https://thunlp.oss-cn-qingdao.aliyuncs.com/ietrans/vg1800_motif_sgcls_I90_E100.pth). The corresponding command is: ```shell bash cmds/1000/motif/demo_sgdet.sh visualization/demo_imgs/ ```
Sorry for the late reply. Please copy the file tools/vg1800/eval.py to your output dir (the dir with eval_results.pytorch). Then, run `python eval.py`. The relational triplet (e.g. top50 triplet: xxxx) results...
感谢关注。这个实时交互的demo整体是一个pipeline系统，由ChatGPT API整合多帧描述，效果确实不太稳定。为了尽量缓解不稳定效果，我们实现的框架思路中大致如下: 1. 尽量让OmniLMM/MiniCPM-V对每帧描述的简短一些，防止过多的无关内容，我们用的prompt是"What is happening in the image?"。 2. 每次回答问题后加了一个历史总结功能来防止上下文过长，使用的prompt是： ``` Before coming to the next round, please make summarization. You need to first illustrate the overall event. For...
Sorry for replying late. We currently do not provide an easy-to-use interface for custom data. You can try to convert it into the Refcoco format. The basic idea is to...
The candidates can be obtained from any object detector, like [Faster-RCNN](https://github.com/rbgirshick/py-faster-rcnn). You can also use the feature extractor of VinVL, which is the prompt_feat's code. To use VinVL detector, you...
Thanks for the interest! The technical report is almost finished. Will release this week.
Which program are you running? Demo or the normal model test? If it is convenient, can you provide more information like the screenshot of the problem and the results you...