pstr_results.pkl

[Ricelll]

hello, i think this is a very outstanding work. I have a question that where i can get pstr_results.pkl? I have install “mmcv” package according to the file of "readme.md"

[Ricelll]

@JialeCao001

[mufasachan]

Hello,

I answer to your question in your reply. Namely, the pkl file is the output of the model. For testing the model we need the output and run the test script. Thus, the test script needs the output of the model, the pkl file.

First, you need to install mmcv correctly in order to run the model. See the installation section of this repo. I advise to run a container to do this, a conda environment should work also. Then get your data and the model weight. Once the pre-requisite are done, you can output the .pkl file, this is binary file with a list of the model outputs. You have to use the run_test_[database].sh, database is cuhk or prw. This script runs 3 Python scripts: the first Python script outputs the .pkl, the second Python script is the "vanilla" evaluation and the third one uses a CBPM (Context Bipartite Graph Matching) to boost the result of the person search task -> better mAP and R1.

[Ricelll]

Hello,

I answer to your question in your reply. Namely, the pkl file is the output of the model. For testing the model we need the output and run the test script. Thus, the test script needs the output of the model, the pkl file.

First, you need to install mmcv correctly in order to run the model. See the installation section of this repo. I advise to run a container to do this, a conda environment should work also. Then get your data and the model weight. Once the pre-requisite are done, you can output the .pkl file, this is binary file with a list of the model outputs. You have to use the run_test_[database].sh, database is cuhk or prw. This script runs 3 Python scripts: the first Python script outputs the .pkl, the second Python script is the "vanilla" evaluation and the third one uses a CBPM (Context Bipartite Graph Matching) to boost the result of the person search task -> better mAP and R1.

thanks very much!!!it's worked out!