From 6a4459ebd7be7dfcb4209316a962f88b68a7a08b Mon Sep 17 00:00:00 2001 From: Brian Davis Date: Tue, 10 Sep 2019 10:16:09 -0600 Subject: [PATCH] Changed ambigious "optimization" references. --- README.md | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/README.md b/README.md index 09b92fd..77a684a 100644 --- a/README.md +++ b/README.md @@ -5,7 +5,7 @@ This is the code from our ICDAR 2019 paper, "Deep Visual Template-free Form Pars This code is licensed under GNU GPL v3. If you would like it distributed to you under a different license, please contact me (briandavis@byu.net). ## Requirements -* Python 3.x (at least 3.7 for doing optimization) +* Python 3.x (at least 3.7 for doing global pairing optimization) * PyTorch 1.0+ I used conda, but it shouldn't be hard to install the packages another way. @@ -26,7 +26,7 @@ Run this in the Visual-Template-Free-Form-Parsing directoty: `python setup.py bu ### CVXPY -To run the optimization, you need cvxpy, which requires python 3.7 or later: +To run the global pairing optimization, you need cvxpy, which requires python 3.7 or later: Clone github: https://github.com/cvxgrp/cvxpy @@ -57,10 +57,10 @@ Detection, full set: `python eval.py -c saved/detector/checkpoint-iteration15000 Detection, pairing set: `python eval.py -c saved/detector/checkpoint-iteration150000.pth -n 0 -T -a data_loader=special_dataset=simple` -Pairing, no optimization: `python eval.py -c saved/pairing/checkpoint-iteration125000.pth -n 0 -T` +Pairing, no pairing optimization: `python eval.py -c saved/pairing/checkpoint-iteration125000.pth -n 0 -T` ` -Pairing, with optimization: `python eval.py -c saved/pairing/checkpoint-iteration125000.pth -n 0 -T -a optimize=true` +Pairing, with pairing optimization: `python eval.py -c saved/pairing/checkpoint-iteration125000.pth -n 0 -T -a optimize=true` #### Perfect information experiments @@ -97,7 +97,7 @@ Scoring functions pairing: `python eval.py -f cf_test_no_vis_pairing.json -n 0 - No visual features pairing: `python eval.py -f cf_test_no_vis_pairing.json -n 0 -T` -No visual features pairing, with optimization: `python eval.py -f cf_test_no_vis_pairing.json -n 0 -T -a optimize=true` +No visual features pairing, with global optimization: `python eval.py -f cf_test_no_vis_pairing.json -n 0 -T -a optimize=true` #### Perfect information experiments @@ -109,7 +109,7 @@ For GT detections: `python eval.py -f cf_test_no_vis_pairing.json -n 0 -T -a useDetect=gt` -For optimization with GT num neighbors: +For global optimization with GT num neighbors: `python eval.py -f cf_test_no_vis_pairing.json -n 0 -T -a rule=closest,optimize=gt` @@ -151,7 +151,7 @@ Evaluating detector: Evaluatring pairing: * `-a useDetect=[gt,path]`: Whether to use GT detections (`gt`) or can be directory with jsons with saved detections. * `-a rule=[closest,icdar]`: Use a rule (nearest or scoring functions) to do pairing (instead of model). -* `-a optimize=[true,gt]`: Use optimization. If `gt` specified it will use the GT number of neighbors. +* `-a optimize=[true,gt]`: Use global pairing optimization. If `gt` specified it will use the GT number of neighbors. * `-a penalty=[float]`: The variable *c* in Equation 1. Default is 0.25 * `-a THRESH=[float]`: Modify the thresh for calculating prec/recall for relationships. Also is *T* in Equation 1. Default is 0.7 * `-a sweep_threshold=true`: Run metrics using a range of thresholds